Sep 2013
No.155
Ethnic Inequality
Alberto AlesinaHarvard University, NBER and IGIER
Stelios Michalopoulos Brown University, NBER and
Elias Papaioannou, London Business School, CEPR and NBER
WORKING PAPER SERIES
Centre for Competitive Advantage in the Global Economy
Department of Economics
Ethnic Inequality∗
Alberto Alesina
Harvard University, NBER and IGIER
Stelios Michalopoulos
Brown University and NBER
Elias Papaioannou
London Business School and CEPR and NBER
First Draft: October 2012
Revised July 2013
Abstract
This study explores the consequences and origins of between-ethnicity economic inequality
both across and within countries. First, combining satellite images of nighttime luminosity
with the historical homelands of ethnolinguistic groups we construct measures of ethnic inequality for a large sample of countries and show that the latter is strongly inversely related to
comparative development. Second, differences in geographic endowments across ethnic homelands explain a sizable portion of ethnic inequality contributing to its persistence over time.
Third, exploiting across-district within-African countries variation using individual-level data
on ethnic identification and well-being from the Afrobarometer Surveys we find that betweenethnic-group inequality is systematically linked to regional under-development. In this sample
we also explore the channels linking ethnic inequality to (under) development, finding that
ethnic inequality maps to political inequality, heightened perceptions of discrimination and
undersupply of public goods.
Keywords: Ethnicity, Diversity, Inequality, Development, Geography
JEL classification Numbers: O10, O40, O43.
∗
We thank Christian Dippel, Nathan Nunn, Debraj Ray, Andrei Shleifer, Enrico Spolaore, Pierre Yared, Romain
Wacziarg, David Weil, Michele Lenza, Oeindrila Dube and Ivo Welch for valuable comments and suggestions. We
also would like to thank for useful feedback seminar participants at Dartmouth, the Athens University of Economics
and Business, UBC, Brown, CREi, Oxford, Bocconi, NYU, Paris School of Economics, Warwick, LSE, Nottingham,
the NBER Summer Institute Meetings in Political Economy, the CEPR Development Economics Workshop, the
conference "How Long is the Shadow of History? The Long-Term Persistence of Economic Outcomes" at UCLA,
and the Nemmers Conference in the Political Economy of Growth and Development at Northwestern University.
All errors are our own responsibility.
0
1
Introduction
Ethnic diversity has costs and benefits. On the one hand, diversity in skills, education, and endowments can enhance productivity by promoting trade and innovation, especially in advanced
economies. On the other hand, ethnic diversity is often associated with poor policies, low public
goods provision, conflict, and hatred. In fact a large literature shows a negative effect of ethnolinguistic fragmentation on various aspects of economic performance, with the possible exception
of wealthy economies (see Alesina and Ferrara (2005) for a review). Income inequality may also
have both positive and negative effects on development. On the negative side, a higher degree of
income inequality may lead to conflict, crime, prevent the poor from acquiring education and/or
lead to expropriation and lofty taxation discouraging investment. On the positive side, however,
income inequality may spur innovation and entrepreneurship by motivating individuals. Further
complicating the relationship between the two, a positive correlation between inequality and development may simply reflect Simon Kuznetz’s conjecture that industrialization translates into
higher levels of inequality at the early stage of development; while at later stages the association
becomes negative. Given the theoretical ambiguities (and data issues) perhaps it comes at no
surprise that it has been very hard to detect empirically a robust association between inequality
and development (see Benabou (2005) and Galor (2011) for surveys).
This paper puts forward and tests an alternative conjecture that focuses on the inter-section
of ethnic diversity and inequality. Our thesis is that what mostly matters for development are
economic differences between ethnic groups coexisting in the same country, rather than the degree of fractionalization per se or income inequality conventionally measured (i.e., independent of
ethnicity). Inequality in income along ethnic lines is likely to lead to political inequality, increase
animosity, and lead to discriminatory policies of one (or more) groups against the others. Furthermore, differences in preferences across both ethnic and income lines may lead to inadequate
public goods provision, as groups’ ideal allocation of public goods will be quite distant. Moreover, the presence of an economically dominant ethnic minority may lower support for democracy
and free-market institutions, as the majority of the population usually feels that the benefits of
capitalism go to just a handful of ethnic groups. Ethnic inequality often implies that well-being
depends on group affiliation; hence it is more likely to generate envy and perceptions that the
system is "unfair", more so than the conventionally measured economic inequality, since the latter
can be more easily be thought of as the result of ability or effort.
The first contribution of this paper is to provide measures of within-country differences in
well-being across ethnic groups, coined as "ethnic inequality". Information on income levels of
ethnic groups for all countries is not available. Hence, to construct country-level indicators of ethnic inequality for the largest possible sample, we combine maps on the location of ethnolinguistic
1
groups with satellite images of light density at night, which are strong proxies of development
and are available at a fine grid (see Henderson, Storeygard, and Weil (2012)). The cross-ethnic
group inequality index is weakly correlated with the commonly employed -and notoriously poorly
measured- income inequality measures at the country level. To isolate the cross-ethnic component of inequality from the overall inequality across regions, we also construct proxies of spatial
inequality.
Second, we document a strong negative association between ethnic inequality and real
GDP per capita at the country level. This correlation holds even when we condition on the
overall degree of spatial inequality, which is also inversely related to well-being (an novel finding
by itself). We also uncover that the negative correlation between ethnolinguistic fragmentation
and development weakens considerably when we account for ethnic inequality; this suggests that
it is the unequal concentration of wealth across ethnic lines that is detrimental for development
rather than diversity per se.
Third, in an effort to shed light on the roots of ethnic inequality we construct measures
reflecting differences in geographic endowments across ethnic homelands and show that the latter
is a strong predictor of ethnic inequality. In contrast there is no link between contemporary ethnic
inequality and historical variables capturing the type of colonization, state formation, legal origin,
etc. Fourth, we show that contemporary development at the country level is also inversely related
to inequality in geographic endowments across ethnic homelands.
Fifth, we examine the link between ethnic inequality and development using individual-level
data from the Afrobarometer surveys, exploiting within-country across-district variation. Besides
the immediate econometric benefits of looking within -rather than across countries- (effectively
accounting for country-level fixed factors related to national institutions and policies, historical
legacies, etc), this is quite useful for validating the cross-country results (based on luminosity
and ethnographic maps) using detailed micro-level data. Specifically, we construct measures
capturing between-group and within-group inequality utilizing information from roughly 20, 000
respondents in 16 Sub-Saharan countries on ethnic identification and well-being. Our analysis
shows that -conditional on numerous individual characteristics- respondents residing in ethnically
unequal districts are less educated and have lower standards of living. This pattern also holds
when we exploit within-ethnicity variation (this is feasible as we observe respondents from the
same ethnicity residing in multiple districts), effectively accounting for both unobserved ethnic
features and migration.
Sixth, we exploit the richness of the Afrobarometer data to make some progress towards
understanding the mechanisms linking ethnic inequality to (lack of) development in the most
unequal and fragmented part of the world, Sub-Saharan Africa. We begin our analysis on the
2
channels showing that ethnic inequality goes in tandem with political inequality and discrimination across ethnic lines. We then examine the link between ethnic inequality and public goods
provision building on the large literature in political economy showing that in ethnically diverse
countries (as well as regions and cities within the United States and some other countries) there
is under-provision of public goods. Our analysis shows that even when one looks within the same
country access to basic public goods, such as piped water, a sewage system, and electrification
is systematically lower in districts characterized by high levels of ethnic inequality. While this
association does not necessarily reflect a causal relationship, it pertains even when controlling
for numerous individual characteristics (including proxies of wealth and well-being) and ethnicity
fixed effects. Finally, we examine the link between ethnic inequality and individual beliefs on the
merits of democracy. This investigation is motivated by the influential thesis of Chua (2003),
who argues that the spread of democracy (and free market institutions) since the 1990s does not
necessarily lead to improved economic efficiency, when wealth is concentrated across ethnic lines.
The uncovered evidence is supportive of this conjecture. Respondents found in ethnically unequal districts feel unfairly treated by the government and are less satisfied with the functioning
of democratic institutions.
Related Works While both the literature linking ethnolinguistic diversity and economic
performance and the literature studying the interplay between inequality and development are
voluminous, there have been very few works examining the role of ethnic inequality on wellbeing. Exploiting data from 29 developing countries (from the Demographic and Health Surveys),
Kyriacou (2013) finds that socioeconomic ethnic group inequalities reduce government quality.
Cederman, Weidman, and Gleditch (2011) combine proxies of local economic activity from the
G-Econ database with ethnolinguistic maps to construct an index of ethnic inequality for a subset of "politically relevant ethnic groups" (as defined by Ethnic Power Relations Dataset) and
then show that in highly unequal countries, both rich and poor groups fight more often than
those groups whose wealth is closer to the country average.
Structure The paper is organized as follows. In section 2 we describe the construction
of the ethnic inequality index and present summary statistics. In Section 3 we report the results
of our analysis associating income per capita with ethnic inequality across 173 countries. In
Section 4 we examine the geographic origins of contemporary differences in ethnic inequality
across countries; we also report estimates associating contemporary development with inequality
in geographic endowments across ethnic homelands. In Section 5 we first examine the withincountry across-district association between ethnic inequality and well-being in 16 Sub-Saharan
countries using individual-level data from the Afrobarometer Surveys. We then use these data to
3
explore the channels linking ethnic inequality to development. In the last section we summarize
and discuss potential avenues for future research.
2
Data and Descriptive Statistics
Since data on income-wages at the ethnicity level are not readily available for a large number of
countries, to construct proxies of ethnic inequality for a large number of countries we combine
information from ethnographic maps on the location of ethnic-linguistic groups with satellite
images on light density at night that are available at a fine grid and can thus be aggregated at
the historical ethnic homeland level. In this Section we discuss the construction of the crosscountry measures capturing inequality in development (as reflected in luminosity) across ethnic
homelands and across regions within 173 countries.
2.1
Location of ethnic groups
We identify the location of ethnic groups employing two data sets/maps. First we use the GeoReferencing of Ethnic Groups (GREG), which is the digitized version of the Soviet Atlas Narodov
Mira (Weidmann, Rod, and Cederman (2010)). GREG portrays the homelands of 1, 276 ethnic
groups around the world. The information pertains to the early 1960’s so for many countries,
in Africa in particular, it corresponds to the time of independence.1 The GIS data set uses the
political boundaries of 1964 to allocate groups to different countries. We thus project the ethnic
homelands to the political boundaries of the 2000 Digital Chart of the World (ignoring polygons
of less than 1 km2 ); this results in 2, 125 ethnic homelands within contemporary countries. Most
areas (1, 630) are coded as pertaining to a single group whereas in the remaining 495 there can be
up to three groups. For example, in Northeast India along an area of 4, 380 km2 the Assamese,
the Oriyas and the Santals overlap. In these cases we assign the respective homeland to all
groups. The size of ethnic homelands varies considerably. The smallest polygon occupies an area
of 1.15 km2 (French in Monaco) and the largest extends over 7, 335, 476 km2 (American English
in the US). The median (mean) group size is 4, 198 (61, 506) km2 . The median (mean) country
has 8 (11.52) ethnicities with the most diverse being Indonesia with 94 groups.
Our second source is the 15th edition of Ethnologue (Gordon (2005)) that maps 7, 570
linguistic groups (using the political boundaries of 2000 for the geo-referencing). In spite of the
detailed linguistic mapping, Ethnologue’s coverage for some continents (e.g., Latin America) is
limited while for others (i.e. Africa and Asia) is very detailed. Ethnologue’s mapping corresponds
to the early 1990’s; thus the location of ethnic groups may be affected by national policies, conflict,
1
The original Atlas Narodov Mira consists of 57 ethnographic maps. The original sources are: (1) ethnographic
and geographic maps assembled by the Institute of Ethnography at the USSR Academy of Sciences, (2) population
census data, and (3) ethnographic publications of government agencies.
4
or other features. Each polygon in the Ethnologue delineates a traditional linguistic region;
populations away from their homelands (in cities, refugee camps) are not mapped. Groups of
unknown location, widespread and extinct languages are not mapped, the only exception is the
English in the United States. Ethnologue also records areas where languages overlap; in this case
we assign the polygon to all languages. Ethnologue provides a more refined linguistic aggregation
compared to the GREG. As a result the median (mean) homeland extends to 728 (12, 986) km2 .
The smallest language is the Domari in Israel which covers 1.18 km2 with the largest group is
the English in the US covering 9, 327, 331 km2 . The median (mean) country has 9 (41.9) groups
with Papua New Guinea being the most diverse country with 791 linguistic groups.
GREG attempts to map major immigrant groups whereas Ethnologue generally does not.
This is important for countries in the New World. For example, in Argentina GREG reports
16 groups, among them Germans, Italians, and Chileans, whereas Ethnologue reports 20 purely
indigenous groups, such as the Toba and the Quechua. For Canada Ethnologue lists 77 mostly
indigenous groups, like the Blackfoot and the Chipewyan with only English and French being historically non-indigenous; in contrast GREG that lists 23 groups is featuring many non-indigenous
groups, such as Swedes, Russians, Norwegians and Germans. Hence, the two ethnolinguistic mappings capture different cleavages, at least in some continents.2
2.2
Luminosity
Since comparable data on income per capita at the ethnicity level across all countries in the
world do not exist, following Henderson, Storeygard, and Weil (2012) and subsequent studies
(e.g., (Chen and Nordhaus (2011), Pinkovskiy (2011), Michalopoulos and Papaioannou (2012,
2013)) we use satellite image data on light density at night as a proxy.3 The luminosity data
come from the Defense Meteorological Satellite Program’s Operational Linescan System that
reports images of the earth at night. The six-bit number that ranges from 0 to 63 is available
approximately at every square kilometer since 1992. To construct luminosity at the desired level
of aggregation we average all observations falling within the boundaries of an ethnic group and
then divide with the population of each area using data from the Gridded Population of the
World that reports geo-referenced pixel-level population estimates for 1990 and 2000.4
2
We are including all groups in our analysis without attempting to make a distinction as to which cleavage is
more salient.
3
These -and other works- show that luminosity is a strong correlate of development at various levels of aggregation (countries, regions, ethnic homelands).
4
The data is constructed using subnational census and other surveys of population at various levels (city,
neighborhood, region). See for details: http://sedac.ciesin.columbia.edu/data/collection/gpw-v
5
2.3
Ethnic Inequality
We proxy the level development in ethnic homeland i with average luminosity per capita, yi , and
construct the Gini index that reflects inequality across ethnic groups (ethnic inequality) for each
country. Specifically, the Gini coefficient for a country’s population consisting of n groups with
values of luminosity per capita for the historical homeland of ethnicity i, yi , where i = 1 to n are
indexed in non-decreasing order (yi ≤ yi+1 ), is calculated as follows:
n
(n + 1 − i)yi
1
G = (n + 1 − 2 i=1n
)
n
i=1 yi
Note that the ethnic Gini index captures differences in mean income -as captured in luminosity per capita at the ethnic homeland- across groups. For each of the two different linguistic
maps we construct Gini coefficients for the maximum sample of countries using cross-ethnichomeland data in 1992, 2000, and 2009. As a robustness we also construct the Gini coefficient
dropping the capital cities and excluding small ethnicities, defined as those capturing less than
1% of the 2000 population in a country. For example, in Kenya the Atlas Narodov Mira (the
Ethnologue) maps 19 (53) ethnic (linguistic) areas. Yet 7 ethnic (37 linguistic) areas are less than
one percent of the Kenya’s population as of 2000. We thus construct the ethnic Gini index using
all ethnic groups (19 and 53), but also just using the 12 large ethnic and 16 large linguistic areas
in Kenya, respectively.
2.4
Spatial inequality
Since we use ethnic homelands (rather than individual-level) data to measure between-group
inequality, the ethnic inequality measures also reflect regional disparities in income (that may
not be related to ethnic diversity). To isolate the between-ethnicity component from regional
inequality, we thus also construct Gini coefficients reflecting the overall (rather than the ethnic)
degree of spatial inequality in each country. Since we couldn’t find a widely-accepted way to
measure spatial inequality, we construct for robustness two measures of the overall degree of
spatial inequality in each country.
Spatial Gini Coefficient 1. This index is based on aggregating luminosity per capita
across roughly equally-sized boxes. We first generate a global grid of pixels of 2.5 by 2.5 decimal
degrees (that extends from −180 to 180 degrees longitude and from 85 degrees latitude to −65
degrees latitude). Second, we intersect the resulting global grid with the 2000 Digital Chart of
the World that portrays contemporary national borders; this results in 4, 512 pixels across the
globe falling within country boundaries. The median (mean) pixel extends to 25, 967 (29, 780)
km2 , being comparable to the size of ethnic homelands in the GREG dataset, when we exclude
those groups with less than 1 percent of a country’s population (20, 338 km2 ). Third, for each
6
pixel (of 2.5x2.5 decimal degrees) we compute luminosity per capita in 1992, 2000, and 2009.
Fourth, we aggregate the data at the country level estimating a Gini coefficient that captures the
overall degree of economic spatial inequality.
Spatial Gini Coefficient 2 Virtual countries created by the 2.5 by 2.5 degree boxes
are on average somewhat larger than ethnic homelands; moreover, because of the fixed grid
dimensionality, smaller countries end up having fewer boxes. Hence, to capture spatial inequality
at a level of aggregation similar to the one in the data we also constructed an index of spatial
inequality based on Thiessen polygons. The latter have the unique property that each polygon
contains only one input point, and any location within a polygon is closer to its associated point
than to the point of any other polygon. Importantly, we use as input points the centroids of the
linguistic homelands according to the Ethnologue dataset. Thus, Thiessen polygons have the exact
same centroid as the actual linguistic homelands in the Ethnologue database; the key difference
being that ethnic homelands rather than being symmetric polygons have idiosyncratic shapes.
We then intersect the 7, 570 Thiessen polygons with the country boundaries in 2000 obtaining a
total of 9, 116 grids. We then construct a spatial Gini coefficient that reflects inequality in lights
per capita across Thiessen polygons.5 The mean size of the Thiessen polygons is 14.809 km, very
similar to the mean size of homelands in the Ethnologue (12, 964 km2 ).
It is important to realize that both proxies of the overall degree of spatial inequality also
reflect inequality across ethnic homelands, since (i) there is clearly some degree of measurement
error on the exact boundaries of ethnic regions and (ii) because population mixing is in practice
higher than the one we observe in the data. Moreover, in countries with large groups the spatial
Gini coefficients may also (partially) capture within-ethnic-group inequality. We thus (almost)
always include both the ethnic inequality and the overall spatial inequality index in the empirical
specifications.6
2.5
Example
Figures 1a − 1b provide an illustration of the construction of the ethnic inequality measures for
Afghanistan. The Atlas Narodov Mira maps 31 ethnicities (Figure 1a). The Afghan is the largest
group that consists of the Pashtuns and the Pathans residing in the southern and central-southern
regions. This group takes up 51% of the population in 2000. The second largest group is the Tajik,
who compose 22% of the population and are located in the north-eastern regions and in scattered
5
To focus on non-trivial grids in terms of size for both the Thiessen polygons and the 2.5 by 2.5 decimal degree
boxes we drop those polygons capturing an area of less than 100 square kilometers.
6
In principle one could generate within-group inequality measures using the finer structure of the luminosity
data. However, within-group mobility and risk sharing issues makes a luminosity-based, within-group inequality
index less satisfactory. We perform a proper decomposition of between and within-group inequality in the last
section using micro-level data from Africa. The micro-level data also allows us to account for migration (as quite
often we observe individuals from the same ethnic groups in more than two regions in a country).
7
pockets in the western part of the country. There are 8 territories in which groups overlap.
We first estimate for each of the 31 homelands luminosity per capita. For groups appearing in
multiple pockets we derive the weighted average of lights per capita using as weights the fraction
of each pocket’s surface area to the total area of the group in the country. Figure 1b portrays the
distribution of lights per capita. Regional development is low in the center of the country, where
the Hazara-Berberi reside and in the eastern provinces, where the Nuristani, the Pamir Tajiks,
the Pashai, and the Kyrgyz tribes are located. Luminosity is higher in the Pashtun/Pathans
homelands and to some lesser extent in the Tajik regions. Second, using lights per capita across
all homelands we estimate the Gini coefficient in 1992, in 2000, and in 2009. In 2000 the Gini
coefficient estimated from GREG is 0.93 very close to the estimate when we use Ethnologue
that maps 39 groups (0.90). We also estimated the ethnic inequality measures excluding the
ethnic homeland where the capital, Kabul, falls; and we also estimated Gini coefficients of ethnic
inequality excluding groups constituting less than 1% of the country’s population.
Figures 2a − 2b illustrate the construction of the overall spatial inequality indicators (Gini
coefficients) using the two different methods. When we divide the globe into pixels of 2.5 x 2.5
decimal-degree boxes we get 24 areas in Afghanistan (Figure 2a). When we use Thiessen polygons
we get 56 pixels in Afghanistan (Figure 2b). After estimating for each pixel, average luminosity
per capita, we aggregate at the country level calculating the Gini coefficient across these pixels.
The resulting measures, overall spatial inequality Gini index 1 and 2 for Afghanistan equal 0.722
and 0.827, respectively.
2.6
2.6.1
Descriptive Analysis
Ethnic Inequality around the World
Table 1 - Panel A reports summary statistics, while Appendix Table 1, Panels A and B report
the correlation structure of the ethnic Gini coefficients between the two global maps in different
points in time. The correlation of the Gini coefficients across the two alternative mappings is
strong, around 0.75 − 0.80. In the relatively short period where luminosity data are available
(1992−2009), ethnic inequality appears very persistent, as the correlations of the Gini coefficients
over time exceed 0.9. Given the high inertia, in our empirical analysis we will exploit the crosscountry variation. The correlation between ethnic inequality and the overall spatial inequality
indicators is high, but far from perfect; ranging between 0.6 − 0.8.
Figures 3a−3d illustrate the global distribution of ethnic and spatial inequality. Africa (and
South Asia) are the most ethnically unequal place(s) in the world. In contrast Western Europe
is the region with the lowest level of ethnic inequality. According to the Atlas Narodov Mira,
the countries with the highest ethnic group inequality are Sudan, Afghanistan, and Mongolia
8
(Gini index higher than 0.90). According to the Ethnologue’s more detailed mapping of ethnic
homelands the countries with the highest cross-ethnic-group inequality (where Gini exceeds 0.90)
are: Chad, Sudan, Papua New Guinea, Brazil, Ethiopia, Angola, Nigeria, Zimbabwe, Zaire,
Cameroon, Laos and Indonesia. The countries with the highest overall spatial inequality in light
density according the measure based on Thiessen polygons (spatial Gini 2 is higher than 0.90)
are Chad, Papua New Guinea, Zaire, Gabon, Congo, the Central African Republic, and Sudan.
Since we are primarily interested in uncovering the explanatory power of ethnic inequality
beyond the overall spatial inequality in most specifications we control for the latter. Figures
3e − 3f portray the global distribution of ethnic inequality partialling out the effect of the overall
degree of spatial inequality. In Figures 4a - 4b we plot ethnic inequality against the overall degree
of spatial inequality. A few interesting patterns emerge. On the one hand, Sudan, Afghanistan,
and Mongolia have much higher ethnic inequality as compared to the overall spatial inequality
(which is also very high). In contrast, USA and Canada score low in ethnic inequality as compared
to the overall degree of spatial inequality (which is high). On the other hand, Azerbaijan, Syria,
Albania, Tunisia, Haiti, and Rwanda score quite high in ethnic inequality, while in contrast the
overall degree of spatial inequality is quite low.
2.6.2
Basic Correlations
Ethnic Diversity Appendix Table 1 - Panel C reports the correlation structure between the
various ethnic inequality and spatial inequality measures with the widely-used ethnolinguistic
fragmentation measures. There is a positive correlation between ethnic inequality and linguisticethnic fractionalization (0.38 − 0.45) (data come from Alesina, Devleeschauwer, Easterly, Kurlat,
and Wacziarg (2003))). In contrast, there is no systematic association between ethnic inequality and religious fractionalization. Figures 5a− 5b provide a graphical illustration (including
continental fixed effects does not change the pattern). The correlation between ethnic inequality and the segregation measures compiled by Alesina and Zhuravskaya (2011) is also positive
(0.20 − 0.45). Ethnic inequality tends to go in tandem with segregation. This is reasonable since
more mixing of groups may lead to a reduction of ethnic-based inequality, which instead is more
likely to persist when groups are geographically separated. We also examine the association between ethnic inequality and spatial inequality with the ethnic polarization indicators of Montalvo
and Reynal-Querol (2005) and Esteban, Mayoral, and Ray (2012), failing to detect a systematic
association. These results show that the ethnic inequality measure captures a dimension distinct
from already-proposed aspects of a country’s ethnic composition.
Income Inequality We then examined the association between ethnic inequality and
income inequality, as reflected in the standard Gini coefficient (Appendix Table 1 - Panel D).
9
The income Gini coefficient is taken from Easterly (2007) who using survey and census data
compiled from the WIDER (UN’s World Institute for Development Economics Research) constructs adjusted cross-country Gini coefficients for more than a hundred countries over the period
1965 − 2000. Figures 6a and 6b illustrate this association using the GREG and the Ethnologue
mapping of ethnic homelands, respectively. The correlation between ethnic inequality and economic inequality is moderate, around 0.25 − 0.30. Yet this correlation weakens considerably and
becomes statistically insignificant once we simply condition on continental constants.
3
Ethnic Inequality and Development
In Table 3 we report cross-country LS estimates associating the log of per capita GDP in 2000,
with ethnic inequality (Appendix Table 1 - Panel D reports the unconditional correlation of
ethnic inequality with various proxy measures of economic and institutional development). In
Panel A we use the ethnic inequality measure using the Atlas Narodov Mira database, while in
Panel B we use the measures derived from Ethnologue’s mapping. In all specifications we include
region fixed effects to account for continental differences in the coverage of ethnic groups and the
huge variation in economic development.
The coefficient of the ethnic inequality index in column (1) is negative and highly significant.
Figures 7a−7b illustrate the unconditional association.7 The estimates in columns (2) and (4) also
reveal a negative association between development and the overall degree of spatial inequality,
as reflected on the Gini coefficient based on pixels of 2.5 by 2.5 degrees and the Gini coefficient
based on Thiessen polygons that have the same centroid as ethnic homelands in the Ethnologue.
This suggests that underdevelopment goes in tandem with regional inequalities.8 In columns (3)
and (5) we include both the ethnic inequality index and the spatial Gini coefficients. The ethnic
inequality index continues to enter with a highly significant estimate that falls only slightly in
absolute value. In contrast the coefficient on the overall spatial inequality drops considerably in
all permutations; moreover the estimate becomes statistically indistinguishable from zero. This
suggests that the ethnic component of regional inequality is the relatively stronger correlate of
underdevelopment.9
In columns (6)-(9) we add the log number of ethnic/linguistic groups in the empirical model.
7
The correlation is somewhat weaker in 2009, 0.60 and 0.51 with the GREG and the Ethnologue maps, respectively; the correlation is somewhat stronger in 1992 (0.67 and 0.60, respectively).
8
This -to the best of our knowledge novel- result is interesting by itself and deserves future work in understanding the inter-play between development and regional disparities in income (see Bolton and Roland (1997) for a
theoretical exposition).
9
There is clearly some degree of measurement error on the exact boundaries of ethnic homelands, while by
construction there is no error on the spatial inequality measures. Additionally, to the extent that populations
mix, the overall spatial inequality index also captures part of ethnic inequality. Both observations suggest that the
coefficient of ethnic inequality on development is likely to be an underestimate of the true magnitudes.
10
In line with previous works, income per capita is significantly lower in countries with many ethnic
(Panel A) and linguistic (Panel B) groups (column (6)); yet the estimates in columns (7)-(9)
clearly show that it is ethnic inequality rather than ethnolinguistic heterogeneity that correlates
with underdevelopment. In columns (10)-(11) we examine whether the significantly negative
association between ethnic inequality and income per capita is driven by an unequal clustering
of population across ethnic homelands; to do so we construct Gini coefficients of population
combining the population estimates in 2000 from the Gridded Population of the World dataset
with the mapping of ethnolinguistic groups. The population Gini index enters with a significantly
negative estimate, implying that under-development is associated with an unequal clustering of
population across ethnic regions. Yet once we include in the specification the ethnic inequality
index and the overall spatial inequality indicators (in (11)), the population Gini coefficient index
turns insignificant. The same applies with the spatial Gini coefficient. In contrast the ethnic
inequality measure retains its economic and statistical significance.
The most conservative estimate on the ethnic inequality index in Panel A (1.08) implies
that a reduction in the ethnic Gini coefficient by 0.25 (approximately one standard deviation,
from the level of Nigeria where the ethnic Gini is 0.76 to the level of Namibia where the ethnic Gini
is 0.50) is associated with a 31% (0.27 log points) increase in per capita GDP. The standardized
beta coefficient of the ethnic inequality index is around 0.20 − 0.30, quite similar to the works on
the role of institutions on development (e.g., Acemoglu, Johnson, and Robinson (2001)).
3.1
Sensitivity Analysis
Other Aspects of the Ethnic Composition In Table 3 we investigate whether other
dimensions of the distribution of the population across groups, related to fractionalization and
polarization, rather than inequality across ethnic lines affect comparative development.
In
columns (1) and (5) we augment the specification with a fractionalization index (from Alesina,
Devleeschauwer, Easterly, Kurlat, and Wacziarg (2003)) whereas in columns (2) and (6) we experiment with Fearon’s (2003) cultural fragmentation index that adjusts the fractionalization index
for linguistic distances among ethnic groups. Doing so has no effect on the coefficient on ethnic
inequality that retains its economic and statistical significance. Moreover, the fractionalization
indicators enter with unstable and statistically insignificant estimates.
Motivated by recent works highlighting the importance of polarization in columns (3), (4),
(7), and (8) we condition on two alternative measures of ethnic polarization (from Montalvo and
Reynal-Querol (2005) and Esteban, Mayoral, and Ray (2012); the latter adjusts for linguistic
differences across groups). Ethnic inequality correlates strongly with development, while the polarization measures enter with insignificant estimates. We also estimated specifications including
11
both the polarization and the fractionalization indicators; in all perturbations the coefficient on
ethnic inequality retains its statistical and economic significance.
Alternative Measures and Geographic Controls In Table 4 we augment the specification with additional controls and experiment with alternative ethnic inequality proxies. In
columns (3), (4), (9), and (10) we use ethnic Gini coefficients that exclude ethnic regions where
capitals fall. In columns (5), (6), (11), and (12) we use ethnic Gini indicators that exclude groups
that constitute less than 1% of a country’s population.10 In all specifications we control for the
overall degree of spatial inequality in lights per capita using the spatial Gini index that is based
on Thiessen polygons and ethnic fractionalization.
In odd-numbered columns we control for a country’s size with the log of population in 2000
and log land area, as ethnic heterogeneity, ethnic inequality, and the overall degree of spatial
inequality are likely to be increasing in size. We also control for the absolute value of latitude,
because development is on average higher far from the equator (e.g., Hall and Jones (1999))
and because diversity is higher in areas close to the equator (e.g., Michalopoulos (2012)). The
ethnic inequality index enters with a negative and significant estimate across all permutations.
In even-numbered columns we condition on a rich set of geographic controls; to avoid concerns of
self-selecting the conditioning set, we follow the baseline specification of Nunn and Puga (2012)
and include (on top of the size controls and latitude) an index of terrain ruggedness, distance
to the coast, an index of gem quality, the percentage of each country with fertile soil and the
percentage of tropical land (the Data Appendix gives detailed variable definitions). The negative
correlation between ethnic inequality and income per capita remains strong. The coefficient
on the ethnic inequality measures is quite similar to the more parsimonious specifications with
the size controls only. Thus while still an unobserved or omitted country-wide factor may jointly
affect development and ethnic inequality, the estimates clearly point out that the correlation does
not reflect (observable) mean differences in geographical characteristics or continental disparities.
Other We performed numerous additional sensitivity checks. For example, we dropped
from the estimation (typically small) countries with just one ethnic (or linguistic) group. The
results remain stable (see Appendix Table 2). We also repeated estimation excluding iteratively
observations from each continent or from each income group (following World Bank’s classification); the results remain intact. We also estimate quantile (and median) regressions to explore
potential heterogeneity in the correlation between ethnic inequality and development.11 The
10
Note that a priori there is no reason in excluding small groups, since ethnic hatred may be directed to tiny
groups that control a significant portion of the economy (Chua (2003)).
11
For example in the similar to column (3)-Table 2a specification, quantile regression estimates are in the range
of −1.0 to −1.95.
12
coefficient on the ethnic inequality index is quite stable across quantiles.
4
4.1
Inequality in Geographic Endowments and Ethnic Inequality
On the Origins of Ethnic Inequality
Given the strong correlation between ethnic inequality and under-development it is intriguing to
examine the roots of ethnic inequality.
We started our exploration of the origins of inequality across ethnic lines by examining the
association between the ethnic inequality proxies and commonly-used historical variables that
have been found to correlate with contemporary development. There is little evidence linking
contemporary differences in ethnic inequality to the legal tradition (La Porta, Lopez-de-Silanes,
Shleifer, and Vishny (1998)), the conditions that European settlers faced at the time of colonization (Acemoglu, Johnson, and Robinson (2001)), the share of Europeans in the population (Hall
and Jones (1999) and Putterman and Weil (2010)), the inclusiveness of early institutions (Acemoglu, Johnson, Robinson, and Yared (2008)), state history (Bockstette, Chanda, and Putterman
(2002)), and borders’ design (Alesina, Easterly, and Matuszeski (2011)); for brevity we report
these results in Appendix Table 3. These insignificant associations suggest that the strong negative correlation between ethnic inequality and development does not reflect the aforementioned
aspects of history.
Then, motivated by the findings of Michalopoulos (2012) that ethnolinguistic diversity
increases with geographic heterogeneity, we conjecture that geographic and ecological endowments
play a role in explaining contemporary differences in income across ethnic lines. To the extent that
land endowments affect the diffusion and adoption of technology, then ethnic-specific inequality in
the distribution of geographic features would manifest itself in contemporary differences in wellbeing across groups. To construct proxies of geographic inequality, we first obtain geo-referenced
data on elevation, land’s suitability for agriculture, distance to the coast and presence of water
bodies (lakes, rivers, and other streams) and construct for each ethnic area the mean value of
each of these measures. We then derive Gini coefficients at a country level that reflect groupspecific inequality in each of these dimensions. Following the same procedure to the one regarding
the construction of spatial inequality in luminosity (see Section 2), we estimated measures of the
overall degree of inequality in geographic endowments, constructing for each of the four geographic
features two spatial Gini coefficients: one based on the 2.5 x 2.5 decimal degrees pixels and one
based on Thiessen polygons.
In Table 5 we explore the association between ethnic inequality and these measures of
inequality in geographic endowments across ethnic homelands. Across all permutations, all four
ethnic Gini coefficients in geographic endowments enter with positive estimates suggesting that
13
ethnic-specific differences in endowments translate into larger contemporary disparities in ethnic
development. Depending on the specification details -GREG or Ethnologue mapping, whether
we use all homelands or drop ethnic regions where capitals fall or small groups, whether we
condition on the level of geography and the overall degree of spatial inequality in each of the
four geographic features- different Gini coefficients of geographic inequality enter with significant
estimates. Thus while we cannot precisely identify which geographic feature(s) matter most,
the message from Table 6 is that differences in geography across ethnic regions translate into
differences in contemporary ethnic inequality.
We thus aggregate the four indexes of ethnic inequality in geographic endowments via
principal components. The use of factor analysis techniques is appealing because we have many
variables (Gini coefficients) that aim at capturing the same concept (with some degree of noise),
in our application inequality in geographic endowments. In line with this, there is strong positive
correlation between the four Gini coefficients (see Appendix Table 4). Table 6 reports the results
of the principal component analysis. The first principal component explains more than half of
the common variance of the four measures of inequality in geographic endowments. The second
principal component explains around 20% of the total variance, while jointly the third and fourth
principal components explain a bit less than a fourth of the total variance. Interestingly, all four
inequality measures load positively on the first principal component. Moreover, the eigenvalue
of the first principal component is close or greater than two (one being the rule of thumb), while
the eigenvalues of the other principal components are less than one. We thus focus on the first
principal component, which given the significant positive loadings of all Gini coefficients, we label
it "inequality in geographic endowments across ethnic homelands".12
In Figures 8a − 8b we plot the ethnic inequality in luminosity against the first principal
component of inequality in ethnic-specific geographic endowments. There is a strong positive
association, suggesting that differences in geography explain a sizable portion of contemporary
differences in development across ethnic homelands.
In Table 7 we formally assess the role of ethnic-specific geographic inequality, as captured by
the composite index of inequality in geographic endowments across ethnic-linguistic homelands on
contemporary ethnic inequality. Columns (1) and (4) show that the strong correlation illustrated
in the figures is not driven by continental differences. In columns (2) and (5) we control for
the overall degree of spatial inequality in geographic endowments augmenting the specifications
with the first-principal component of the Gini coefficients in geography (using Thiessen polygons
with the same centroid as ethnic homelands). This has little effect on the coefficient of the
12
We also estimated and incorporated in the factor analysis inequality across ethnic homelands on distance to the
capital city and on the presence of natural resources (namely oil, gold, and diamonds). The results, however, are
similar, in the sense that the first principal component is quite similar and adding more geographic Gini indicators
does not affect much the estimates.
14
ethnic inequality in geographic endowments that retains its economic and statistical significance.
In contrast the Gini coefficient based on Thiessen polygons that captures the spatial degree in
geographic inequality enters with a small and statistically insignificant estimate. In columns (3)
and (6) we control for the level effects of geography, augmenting the specification with mean
elevation, land area under water, distance to the coast, and land suitability for agriculture. In all
permutations the composite index reflecting differences in geographic endowments across ethnic
homelands enters with a positive and highly significant coefficient.13 The estimate in column
(3) implies that a one-standard-deviation increase in the inequality in geography across ethnic
homelands index (1.56 points, say from Mozambique to Malawi) translates into an 15 percentage
points increase in the ethnic inequality index (somewhat more than half a standard deviation;
see Table 1A).
4.2
Geographic Inequality and Development
Given the strong positive association between ethnic inequality -as reflected in lights per capita
across ethnic homelands- and inequality in geographic endowments, it is interesting to examine
whether contemporary development is systematically linked to the unequal distribution of geographic endowments across ethnic homelands. We thus estimated LS specifications associating the
log of real GDP p.c. in 2000 with the composite index of ethnic-specific inequality in geography.
While omitted-variables concerns cannot be eliminated, examining the role of inequality in geographic endowments across ethnic homelands on comparative development is useful in assuaging
concerns that the estimates in Tables 2 − 4 are driven by reverse causation. Moreover, geographic
inequality can be thought of as an alternative "primitive" measure of economic differences across
linguistic homelands (compared to the ethnic inequality index based on luminosity).
Results Table 8 reports the results. The coefficient on the proxy of ethnic inequality in
geographic endowments in (1) and (4) is negative and highly significant suggesting that countries
with sizable inequalities in geographic endowments across ethnic homelands are less developed.
In columns (2) and (5) we condition on the overall degree of inequality in geography with the
spatial Gini index based on Thiessen polygons, while in (3) and (6) we also control for land quality,
elevation, land area under water, and distance to the coast. The coefficient on the inequality in
geographic endowments across ethnic homelands index is negative in all permutations. The
coefficient is statistically different than zero in all but one specifications. In contrast the estimate
on the principal component that reflects the overall spatial inequality in geographic endowments
13
Appendix Table 5 shows that the results are similar when we exclude from the estimation ethnic regions where
capital cities fall and small ethnic groups consisting less than 1% of a country’s population.
15
is quantitatively small, changes sign and is statistically insignificant.14 The estimates in columns
(1) and (3) imply that a one-standard-deviation increase in geographic inequality across ethnic
homelands (1.5 points) decreases income per capita by approximately 30% (0.27 log points).
These results further show that inequality across ethnic regions is a feature of under-development.
Further Evidence We also estimated two-stage-least-squares estimates associating geographic inequality across ethnic homelands to ethnic inequality in lights per capita in the firststage and the component of ethnic inequality explained by geographic disparities across ethnic
regions with log per capita GDP in 2000 in the second stage. While the 2SLS estimates do not
necessarily identify the causal effect of ethnic inequality on development, they may be useful in
accounting for measurement error in the proxy measure of development (lights per capita). The
results (reported in Appendix Table 7) show the 2SLS estimate on the ethnic Gini coefficient is
highly significant and quite similar in magnitude to the LS estimate.
We also estimated specifications linking development to both the ethnic inequality measure
(based on lights per capita) and the composite index capturing inequality in geographic endowments across ethnic homelands. The results, shown in Appendix Table 8, show that once we
condition on contemporary ethnic inequality differences in endowments across ethnic homelands
lose their explanatory power. While some peculiar type of measurement error may explain this
finding, this result indicates that inequality in geographic endowments across ethnic homelands
affects contemporary development via its role on ethnic inequality.
Discussion The results in this section should not be interpreted as proving that unequal
geography across ethnic lines necessarily "causes" ethnic inequality (and under-development). It
is possible that certain groups for a plethora of reasons (e.g., faster early development, superior
military technology, or genetic differences) conquered higher quality territories. In this regard the
correlation between inequality in geographic endowments across ethnic lines and ethnic inequality in development (captured by lights per capita) indicates the sizable persistence of inequality.
Hence, one might view an unequal ethnic geography as a manifestation of deeper ethnic differences. Nevertheless, even in this case, it is the presence of an inherently unequal geography in
a country that partially allows these primordial ethnic differences to become salient (otherwise
there would be no "better land" for stronger groups to conquer and every group would have the
same land endowment).
14
Appendix Table 6 reports otherwise identical specifications using the inequality measures that exclude from
the estimation capitals and small ethnic groups. The results are similar.
16
5
Individual-Level Evidence from Sub-Saharan Africa and a Primer
on the Channels
In this section we take a micro approach that explores within-country across-district variation
in ethnic inequality and various development outcomes in a sample of 16 Sub-Saharan African
countries. The analysis serves two purposes. First, it is quite useful exploring the association between inequality across (and within) ethnic groups and well-being using micro data and exploiting
within (rather than across) country variation. Hence, instead of assigning parts of a country to
a single (or more) groups with linguistic maps, we use self-reported data on ethnic identity and
living conditions minimizing measurement error on the construction of ethnic inequality measures
and also accounting for migration (as we observe people from the same ethnic group in many
regions). Second, the plethora of questions of the Afrobarometer Surveys allows us to shed light
on the channels that link ethnic inequality to well-being.
Our focus on Sub-Saharan Africa is natural, as Africa is by far the most ethnically diverse
part of the world, while ethnic inequality is also quite high. Moreover, previous works suggest that
a considerable portion of Africa’s growth tragedy may be attributed to its ethnic diversity and
patronage politics across ethnic lines (e.g., Easterly and Levine (1997), Franck and Rainer (2012)).
In the same vein, the literature on the origins of African political and economic development mostly in political science- places a key role to ethnic disparities in income (e.g., Robinson (2001)).
5.1
5.1.1
Ethnic Inequality and Well Being
Data
We use individual-level survey data from the 3rd round of the Afrobarometer surveys, that cover
16 Sub-Saharan African in 2005.15 The surveys are based on interviews of a random sample of
either 1, 200 or 2, 400 individuals in each country. We consider all individuals that have a clearly
identified ethnic identity and answer the questions on individual well-being.16 This is the case
for 20, 617 out of 25, 200 respondents; individuals reside in 1, 298 districts. Overall there are 213
ethnic groups across the 16 countries. In each district there are on average 3 ethnic groups (range
from 1 to 23 ethnicities). A (nice) feature of the data -that we exploit below- is that individuals
from the same ethnic group are present in more than one districts.
We construct inequality measures at the district level using individual responses on an
ordered (1−5) living conditions index where a score of 1 indicates very bad conditions; 2 fairly bad;
15
These countries are: Benin, Botswana, Ghana, Kenya, Lesotho, Madagascar, Malawi, Mozambique, Mali,
Namibia, Nigeria, Senegal, South Africa, Tanzania, Uganda, and Zambia. The 3rd round of the Surveys was also
conducted in Cape Verde and Zimbabwe but in these countries the ethnic identification question was not asked.
16
We also drop individuals from ethnic groups with less than 10 respondents in each country (as in this case
both the ethnic-specific mean and the variance of living conditions are likely to be rather imprecisely estimated).
17
3 neither good nor bad; 4 fairly good, and 5 very good. Based on these data we calculate Theil
indicators reflecting the between-group and the within-group components of overall inequality
(the results are similar with the Gini index; we prefer the Theil index as it can be decomposed to
a between ethnic group and a within ethnic group component). Table 1-Panel B reports summary
statistics of the overall, the between, and the within ethnic group inequality measures as well as
the other measures we consider below.
The formulae below describe the between-group and the within-group components of inequality in district d. Namely, for n groups within a district, yi,j denotes the living conditions of
individual j belonging to group i; yi reflects the mean living conditions of group i, and Y denotes
the total living conditions in a given district; ni is the number of individuals belonging to group
i and N is the total number of individuals in the district.
Tbetween =
5.1.2
yi
i Y
ln(
yi
Y
ni
N
)
Twithin =
yi
i Y
yi,j
j yi
ln(
yi,j
yi
1
ni
)
District-Level Estimates
Table 9 reports the results of the within-country analysis that associates ethnic inequality at
the district-level with well-being, as reflected in the 1 − 5 living conditions index (in Panel A)
and a 0 − 9 education index (in Panel B). We augment the specification with the within-ethnicgroup Theil index, so as to jointly examine the role of between and within-group inequality.
In all specifications we control for the log number of ethnicities and the log of the number of
respondents in each district (the Data Appendix gives variable definitions and sources for all
variables, Table 1-Panels B and C report summary statistics while Appendix Table 9 provides
the pairwise correlations). Odd-numbered columns report estimates across all districts, while in
even-numbered columns we drop districts with only one ethnicity (as in these cases the betweenethnic-group inequality is zero by construction).
Specifications (1) and (2) in Panel A show that district-level development -as reflected in
the average living condition across respondents- is significantly lower in districts with high levels
of between-group inequality. The coefficient on the between-group inequality index implies that
a 5% decrease in the Theil index (approximately two standard deviations; see Table 1-Panel B)
is associated with a 0.31 point increase in the average living conditions in a region (close to half
a standard deviation). The coefficient on the within-ethnicity Theil index is also negative and
highly significant, suggesting that inequality within groups is also a feature of regional underdevelopment. The estimate on the within-group Theil index suggests that a 5% decrease in
inequality (a bit more than one standard deviation) is associated with a 0.27 points increase
in the average level of living conditions in a district. The results are similar when we use the
average level of education at the district level as the dependent variable (in Panel B). Regions
18
inhabited by more educated respondents are characterized by a lower degree of both between and
within-group disparities.
5.1.3
Individual-level Estimates
In columns (4) to (6) we report specifications associating between-group and within-group inequality at the district level with living conditions (Panel A) and education at (Panel B) at the
individual level. Moving to the individual level allows us to condition on numerous individual
characteristics (Table 1C reports summary statistics). Following Nunn and Wantchekon (2011)
we control for the respondent’s age and age squared, a gender indicator, a dummy variable indicating urban households, 22 religion fixed effects, and 25 occupational constants. We also control
for the share of the district’s population that is of the same ethnicity as the respondent. The estimates on the between and the within ethnic inequality measures in columns (3) and (4) suggest
that the negative correlations with the development outcomes (living conditions and education)
are not driven by (observable) individual characteristics.
Since members of the same ethnic group are present in more than one district, we also
explore whether conditional on ethnic-specific (observable and unobservable) characteristics, inequality across ethnic lines is an important correlate of individual well-being and education.
Overall almost all (210 out of the 213) ethnic groups in our sample may be found in more than
one district. The median ethnicity can be found in 20 districts. Conditioning on ethnicity fixed
effects seems a priori important, because recent works show that ethnic-specific historical traits,
related, for example, to the slave trades (e.g. Nunn (2008)), pre-colonial political centralization
(e.g., Gennaioli and Rainer (2007) and Michalopoulos and Papaioannou (2013)), and ethnic partitioning (e.g., Michalopoulos and Papaioannou (2011)) have long-lasting effects on development.
The inclusion of country-ethnicity fixed effects also ensures that the negative association between ethnic inequality and development is not driven by certain ethnic groups -that may either
dominate politics in one country or suffer from discrimination.
The estimates on columns (5) and (6), imply that conditional on an array of individual
characteristics, respondents from the same ethnic group report worse living conditions when they
reside in districts characterized by larger ethnic inequality; also individuals in ethnically unequal
regions are also less educated. The estimate in column (6) in Panel B on the between-group Theil
index (−3.03) suggests that a one standard deviation increase in ethnic inequality is associated
with 0.25 standard deviation drop in the level of education. Interestingly within-group inequality
is also negatively related to both living conditions and education; the coefficient on the within
ethnic inequality Theil in column (6) of Panel B suggests that a one standard deviation increase
in the within-group inequality measure is associated with a 0.42 drop in education.
19
Sensitivity Analysis We perturbed the empirical model in various ways to explore the
robustness of these results. First, rather than conditioning on the log number of ethnicities we
constructed a standard fractionalization measure. This has no effect on our main results. Second,
we repeated estimation using the mean log deviation index or the Gini coefficient. The results are
virtually unchanged. Third, we repeated the analysis restricting estimation to urban districts.
This is useful as inequality is higher in urban places and, unlike the cross-country setting, we can
properly account for increased population mixing usually associated with urbanization. Across all
permutations, the coefficient on the between-ethnic-group Theil index is negative and significant
at the 99% confidence level. Fourth, since the living conditions and the education measures
take discrete values we estimated ordered probit models with maximum likelihood again finding
similar results. To avoid cluttering the exposition the results are available upon request.
Summary To the best of our knowledge the results in Table 9 are the first piece of
evidence showing in a systematic manner with micro level data that inequality, both across
ethnic lines and within ethnicities, is inversely related to economic well-being in Sub-Saharan
Africa.
5.2
Channels
We now exploit the richness of the questionnaires of the Afrobarometer Surveys to explore several
channels linking ethnic inequality and well-being.
5.2.1
Economic Conditions and Political Power
Given the large amount of evidence on ethnic-based patronage politics and discrimination in
Africa, we begin our analysis on the mechanisms linking ethnic inequality to under-development
examining the link of a group’s well-being and its political power. Economic differences across
groups may translate into differences in the influence that ethnicities exert on the political landscape (and vice versa). We test this conjecture tabulating respondents’ answers to a question
aiming to measure the political influence of their group to that of other groups in the same country; the question gauging the political influence of the group reads: "Think about the conditions
of [respondent’s identity group]. Do they have less, the same, or more influence in politics than
other groups in this country? " Higher values in this index -that ranges from 1 to 5- indicate more
influence (see the Data Appendix for details).
Figure 9a portrays the relationship between the average of a group’s mean living conditions
and its perceived political leverage (partialling out country-specific constants and the size of each
group). The evident positive association indicates that the economic standing of a group is
closely linked to its political power. Interestingly, the standardized "beta" coefficient of a group’s
20
mean living conditions (0.36) is quite similar to the (similarly positive) "beta" coefficient of
the log number of respondents of the group (0.43), implying that the role of economic wellbeing on political power is similar to that of group size; the latter is consistent with the recent
findings of Francois, Rainer, and Trebbi (2013) who document that a group’s size determines its
representation in the government across most Sub-Saharan African countries.
A similar picture emerges when we look at how a group’s living conditions relate to whether
the group’s respondents feel discriminated by the government; we derive a proxy of group discrimination using the average value to the following question: "How often are [respondent’s identity
group] treated unfairly by the government? " The index ranges from 0 to 3 with higher values
indicating a higher frequency of unfair treatment. Figure 9b shows a strong association between
group-specific living conditions and group’s unfair treatment by the central government.
We also experimented with alternative measures of a group’s economic conditions. Specifically, instead of measuring group’s economic status by taking the average of living conditions
across individuals, we use information from another question that reflects perceptions of group
members regarding the economic power of their own group relative to other groups in the country.
The question reads: "Think about the condition of [respondent’s identity group]. Are their economic conditions worse, the same as, or better than other groups in this country? " Higher values
(range 1 to 5) indicate higher economic prosperity. There is a strong link between group-specific
economic conditions (versus other groups) and political influence or (lack thereof) discrimination
by the government. Please see Figures 10a, 10b, 10c, respectively.
5.2.2
Public Goods Provision
The Argument A large literature provides compelling evidence that public goods pro-
vision, redistributive policies, and effective governance are less prevalent in ethnically/racially diverse communities (e.g., Alesina, Glaeser, and Sacerdotte (2001)) and countries (Desmet, OrtuñoOrtín, and Wacziarg (2012)). The more dispersed preferences are across groups then the desired
public goods will be more distant to the chosen one (see Luttmer (2001) for evidence). Hence, if
the level of income shapes preferences for public goods then group differences in economic conditions will make group preferences diverge leading to lower public goods provision and increased
political tensions.17 In the same vein, one of the main empirical patterns of urban economics is
that the rich quite often want to "isolate"; since this desire may be especially strong when wealth
17
Baldwin and Huber (2010) provide empirical evidence in line with this idea for 46 democracies. Similarly,
Deshpande (2000) and Anderson (2011) focus on income inequality across castes in India and associate betweencaste inequality to public goods. See also Loury (2002) for an overview of works studying the implications of the
evolution of racial inequality in the US.
21
is correlated with group identity, ethnic inequality may lead to segregation.18 The link between
political inequality and economic inequality (shown above) may limit the supply of public goods
to geographically isolated minorities, with an insufficient supply to poorer and less powerful
groups. In contrast to the least well-off, the "rich" may get the desired services (of public goods)
by directly purchasing the respective private goods; obtaining, for example, a private electricity
generator instead of contributing to the extension of the public grid. This phenomenon would
exacerbate the under-provision of public goods.
Results Table 10a presents results relating inequality to access to various type of public
goods, namely piped water, sewage systems and electrification at the enumeration area. The unit
of analysis in columns (1), (4) and (7) is the district, while in the other columns the unit of analysis
is the individual. To further account for individual features, following Nunn and Wantchekon
(2011) in these specifications we also include as additional controls five dummies for household’s
living conditions and nine indicators for education. To economize on space we focus on districts
with more than one ethnic group (the pattern is the same for the full sample as already shown
in Table 9). In all specifications the between-group inequality measure enters with a significantly
negative sign. On the contrary, the explanatory power of within-group inequality index is small
and the coefficient is often statistically indistinguishable from zero. For example, in column (5)
the beta coefficients for the between and within-group inequality are 0.049 and 0.006, respectively.
The results in Table 10a thus show that in districts marked by sizable inequalities across ethnic
lines, there is under-provision of public goods. Since we do not have random assignment on
ethnic inequality, these estimates do not necessarily identify causal effects; yet it is important
to note that we control for numerous individual level characteristics, including education, living
conditions, occupation, and in many specifications even for ethnicity fixed factors.
In Table 10b we report otherwise identical to Table 10a specifications but we now include as
an additional regressor the mean value of the living conditions index in each district. (Note that
in the individual-level regressions we continue controlling for individual level living conditions and
education). Access to public goods is negatively related to between-group inequality; in contrast
the correlation between public goods provision and within-group inequality is weak. This suggests
that the effect of access to public goods goes above and beyond the effect of ethnic inequality on
average well being, since we are controlling for it in these regressions.
Examples A few examples are useful to illustrate the pattern. We focus on the most
restrictive specifications that include ethnicity fixed-effects. The Pular in Senegal are found in 28
18
Alesina and Zhuravskaya (2011) show that ethnic and linguistic segregation correlate negatively with proxies
of effective governance.
22
of the 31 country’s districts. In the district of Matam, where the Pular coexist with the Soninke,
the Wolof and the Mandinka, between-group inequality is minimal (0.0005) whereas in the district
of Sedhiou, where the Pular coexist with the Wolof, the Mandinka, the Manjack, the English, the
Diola, and the Bambara, between-group inequality is 0.0145. In the Sedhiou district all Pular
report having no access to electricity, piped water and sewage system, whereas in Matam 72%
of the Pular have access to an electricity grid and access to clean water. Another example is
the Herero that are found in 32 of the 87 districts in Namibia. In the district of Otjiwarongo,
where we observe respondents from 5 groups, between-group inequality is minimal, 0.0038. In
this region all Herero reply having access to electricity, sewage system, and clean water. On the
contrary, in the district of Otjinene where between-group inequality is more than 10 times larger,
0.0408, only 43% of Herero reply having access to either an electricity grid or a sewage system
(in both regions within-group inequality is quite similar).
5.2.3
Fairness, Markets and Democracy
The argument In an influential work, Chua (2003) notes cases in which the spread of
democracy and free market institutions in the 1990s has led to animosity and institutional capture
by amplifying pre-existing ethnic tensions. Her thesis is that in the presence of ethnic inequality
(and especially when a small fraction of the population controls most wealth in the economy),
then democratization may result in a backlash, as most individuals perceive representative and
capitalist institutions as unfair, captured, and corrupt.
19
When the less-privileged, but more populous groups, come to power they may want to turn
the cards around, pursuing ethnic politics aiming to compensate their group for the perceived
injustice. The resulting belief that markets are unfair may also lead to under-supply of effort (as in
model by Alesina and Angeletos (2005) and Benabou and Tirole (2006)). The strong correlation
documented above between well being and political influence goes in the same direction. It
undermines the perception of "equity" in the society.
Results Tables 11a and 11b report the results of our analysis examining the link between ethnic inequality (and within-ethnic-group) inequality and perceptions on the functioning
of democratic institutions. In columns (1)-(3) the dependent variable captures whether the respondent feels that in some instances a non-democratic government is preferable to a democratic
one. In columns (4)-(6) the dependent variable is a 0 − 4 range index reflecting the degree of satisfaction with the way democracy works in the respondent’s country, whereas in columns (7)-(9)
19
Chua (2003) presents case-study evidence supporting her thesis. She discusses, among others, the influence
of Chinese minorities in Philippines, Indonesia, and other Eastern Asian countries; the dominant role of (small)
Lebanese communities in Western Africa and the similarly strong influence of Indian societies in Eastern Africa.
Other examples, include the I(g)bo in Nigeria and the Kikuyu in Kenya.
23
the dependent variable is an ordered variable (range from 1 to 5) where reflecting respondents’
belief on whether the constitution expresses the values and hopes of people. (So in all specifications higher values in the dependent variable reflect greater support/satisfaction with the
democracy/constitution). To get an idea of the overall variation in the outcome of interest 23%
of the respondents either (strongly) believe that the constitution does not reflect the values of
the people of the country with a similar 19% answering the opposite.
The evidence in Table 11a reveals that support and especially satisfaction with democracy is overall negatively related to across-group inequality and to some extent to within-group
inequality. It is worthwhile to note that amongst the individual controls we always include the
living conditions and education level of the respondent, hence the negative association between
ethnic inequality and support for democracy operates beyond the influence of the former on the
well-being of the respondent. In Table 11b we also control for the overall level of well being of
the district as a whole. Doing so weakens considerably the role of cross-group inequality and
eliminates any influence of the within-group component. This finding suggests that the link
between ethnic inequality and disaffection with democratic institutions works primarily (though
not entirely) through the relationship of ethnic inequality and district-level wellbeing documented
above.
5.2.4
Other Channels
We also examined the link between ethnic inequality and proxies of social-civic capital investigating whether ethnic inequality is related to trust. The results are inconclusive. It seems that
in Africa at least ethnic inequality and trust are not systematically linked.
Finally, we correlated inequality across and within ethnic groups with proxies of conflict at
the regional level, using geo-referenced data on conflict from the ACLED database. While many
have linked (ethnic and overall) inequality with conflict (see Horowitz (1985) for an overview),
as shown theoretically by Esteban and Ray (2011) and Esteban, Mayoral, and Ray (2012) the
relationship is not straightforward; for instance, a very rich and powerful group may be so strong
that any insurrection from (much) weaker groups is unthinkable. Likewise, a change in ethnic inequality starting from a relatively equal equilibrium may generate violence. Given the theoretical
ambiguities, perhaps it comes at no surprise that we could not detect a systematic relationship
between ethnic (or within group) inequality and conflict.20
20
To economize on space, we do not report these results, which are available upon request.
24
6
Conclusion
This study shows that ethnic differences in economic performance rather than the degree of
diversity or the overall level of inequality are negatively correlated with economic development.
While a large literature has examined (a) the interplay between inequality and development and
(b) the effects of various aspects of the ethnic composition (such as fragmentation, polarization,
segregation) on economic performance, there is little -if any- work studying the linkages between
ethnicity, inequality, and comparative development. This paper is a first effort to fill this gap.
In the first part of the paper we examine the role of ethnic inequality on development using
cross-country data. First, combining linguistic maps on the spatial distribution of ethnic groups
within countries with satellite images of light density at night we construct Gini coefficients
reflecting inequality in well-being across ethnic lines for a large number of countries. Ethnic
inequality is weakly correlated with the standard measures of income inequality and only modestly
correlated with ethnolinguistic fractionalization, polarization, and segregation. Second, we show
that the newly constructed proxy of ethnic inequality is strongly negatively correlated with per
capita GDP across countries. The correlation retains its significance when we condition on
the overall degree of spatial inequality in development, which is also negatively associated with
economic development (a new finding by itself). Including in the empirical specification both the
ethnic inequality index and the widely-used ethnolinguistic fragmentation indicators, the latter
loses significance, suggesting that it is inequality across ethnic lines that is correlated with poor
economic performance rather than fractionalization. Third, we conduct an initial step exploring
the roots of contemporary differences in well-being across ethnic groups within countries. In this
regard, we construct indicators of ethnic inequality in geographic endowments and show that
contemporary differences in development across ethnic homelands have a significant geographic
component. The latter is also inversely related to contemporary development.
In the second part of the paper we take a micro approach exploiting within-country differences in ethnic inequality for 16 Sub-Saharan countries covered by the Afrobarometer Surveys.
First, using individual-level data on ethnic self-identification and well-being we find a negative
association between ethnic inequality and regional development. The negative correlation between ethnic inequality and well-being retains significance even when we control for numerous
individual characteristics and account for unobserved ethnic-specific traits. Second, we explore
the mechanisms linking ethnic inequality to regional development. We show that economic inequality goes in tandem with political inequality. Our analysis further establishes that ethnic
inequality is closely linked to the under-provision of basic public goods and to some extent to the
lack of confidence in democratic institutions.
Overall, we view our work as a first step towards mapping and understanding the conse25
quences and origins of contemporary differences in income across ethnic groups. Future research
should further explore the channels via which ethnic inequality interacts with comparative development. Moreover, theoretical work is needed to shed light on the interaction between ethnic
identity, inequality, and various aspects of economic efficiency. We plan on tackling some of these
questions in future work.
26
References
A, D., S. J, J. A. R (2001): “The Colonial Origins of Comparative Development: An Empirical Investigation,” American Economic Review, 91(5), 1369—1401.
A, D., S. J, J. A. R, P. Y (2008): “Income and Democracy,” American Economic Review, 98(3), 808—842.
A, A., G.-M. A (2005): “Fairness and Redistribution: US vs Europe,”
Ameican Economic Review, 95(4), 960—980.
A, A., A. D, W. E!, S. K, R. W# (2003):
“Fractionalization,” Journal of Economic Growth, 8(2), 155—194.
A, A., W. E!, J. M#$ (2011): “Artificial States,” Journal of the
European Economic Association, 9(2), 246—277.
A, A., E. L. F (2005): “Ethnic Diversity and Economic Performance,”
Journal of Economic Literature, 43, 762—800.
A, A., E. G, B. S (2001): “Why Doesn’t the United States Have
a European Welfare State?,” Brookings Papers on Economic Activity, 2, 187—278.
A, A., E. Z$! (2011): “Segregation and the Quality of Government in a
Cross-Section of Countries,” American Economic Review, 101(5), 1872—1911.
A, S. (2011): “Caste as an Impediment to Trade,” American Economic Journal: Applied Economics, 3(1), 239—263.
B, K., J. D. H (2010): “Economic versus Cultural Differences: Forms of
Ethnic Diversity and Public Goods Provision,” American Political Science Review, 104(4),
644—662.
B, R. (2005): “Inequality, Technology, and the Social Contract,” in Handbook of Economic Growth, ed. by P. Aghion, and S. Durlauf, vol. 1, chap. 25, pp. 1595—1638. Elsevier,
North-Holland, Netherlands.
B, R., J. T (2006): “Belief in a Just World and Redistribute Politics,”
Quarterly Journal of Economics, 121(2), 699—746.
B$, V., A. C, L. P (2002): “States and Markets: The Advantage of an Early Start,” Journal of Economic Growth, 7(4), 347—369.
27
B, P., G. R (1997): “The Breakup of Nations: A Political Economy Analysis,”
Quarterly Journal of Economics, 112(4), 1057—1090.
C, L.-E., N. B. W, K. S. G (2011): “Horizontal Inequalities
and Ethnonationalist Civil War: A Global Comparison,” American Political Science Review,
105(1), 478—495.
C, X., W. D. N (2011): “Using Luminosity Data as a Proxy for Economic
Statistics,” Proceeedings of the National Academy of Sciences, 108(21), 8589—8594.
C, A. (2003): World on Fire: How Exporting Free-Market Democracy Breds Ethnic Hatred
and Global Instability. William Heinemann, London, UK.
D/, A. (2000): “Does Caste Still Define Disparity? A Look at Inequality in Kerala,
India,” American Economic Review, Papers and Proceedings, 90(2), 322—325.
D, K., I. O2-O3, R. W# (2012): “The Political Economy of Ethnolinguistic Cleavages,” Journal of Development Economics, 97(2), 322—338.
E!, W. (2007): “Inequality Does Cause Underdevelopment,” Journal of Development
Economics, 84(2), 755—776.
E!, W., R. L (1997): “Africa’s Growth Tragedy: Policies and Ethnic Divisions,” Quarterly Journal of Economics, 112(4), 1203—1250.
E, J., L. M!, D. R! (2012): “Ethnicity and Conflict: An Empirical Study,”
American Economic Review, 102(4), 1310—1342.
E, J., D. R! (2011): “Linking Conflict to Inequality and Polarization,” American
Economic Review, 101(4), 1345—1374.
F$, R., I. R (2012): “Does the Leader’s Ethnicity Matter? Ethnic Favoritism,
Education and Health in Sub-Saharan Africa,” American Political Science Review, forthcoming.
F, P., I. R, F. T (2013): “How Is Power Shared In Africa?,” Working
Paper, NBER, No. 18425.
G, O. (2011): “Inequality, Human Capital and the Process of Development,” in Handbook
of the Economics of Education, ed. by S. M. Eric A. Hanushek, and L. Woessmann. Elsevier,
Amesterdam, Holland.
28
G, N., I. R (2007): “The Modern Impact of Precolonial Centralization in
Africa,” Journal of Economic Growth, 12(3), 185—234.
G, R. G. (2005): “Ethnologue: Languages of the World,” .
H, R. E., C. I. J (1999): “Why Do Some Countries Produce So Much More Output
Per Worker Than Others?,” Quarterly Journal of Economics, 114(1), 83—116.
H, V. J., A. S!, D. N. W (2012): “Measuring Economic Growth
from Outer Space,” American Economic Review, 102(2), 994—1028.
H, A., R. S, B. A (2011): “Penn World Table Version 7.0,” Center for
International Comparisons of Production, Income and Prices at the University of Pennsylvania.
H#, D. L. (1985): Ethnic Groups in Conflict. University of California Press, Berkeley
and Los Angeles, California.
K4, D., A. K!, M. M## (2005): “Governance Matters IV : Governance
Indicators for 1996-2004,” Policy Research Working Paper Series, The World Bank.
K!, A. P. (2013): “Ethnic Group Inequalities and Governance: Evidence from Developing Countries,” Kyklos, 66(1), 78—101.
L P, R., F. L/#--S, A. S4, R. V! (1998): “Law and Finance,” Journal of Political Economy, 106(6), 1113—1155.
(1999): “The Quality of Government,” Journal of Law Economics and Organization,
15(1), 222—279.
L!, G. C. (2002): The Anatomy of Racial Inequality. Harvard University Press, Cambridge,
MA.
L, E. F. (2001): “Group Loyalty and the Taste for Redistribution,” Journal of Political
Economy, 109(3), 500—528.
M/, S. (2012): “The Origins of Ethnolinguistic Diversity,” American Economic
Review, 102(4), 1508—1539.
M/, S., E. P/ (2011): “Divide and Rule or the Rule of the Divided? Evidence from Africa,” NBER Working Paper 17184.
(2012): “National Institutions and Sub-national Development in Africa,” NBER Working Paper 18275.
29
(2013): “Pre-colonial Ethnic Institutions and Contemporary African Development,”
Econometrica, 81(1), 113—152.
M, J.-G., M. R!-Q (2005): “Ethnic Polarization, Potential Conflict,
and Civil Wars,” American Economic Review, 95(3), 796—816.
N, N. (2008): “The Long Term Effects of Africa’s Slave Trades,” Quarterly Journal of
Economics, 123(1), 139—176.
N, N., D. P (2012): “Ruggedness: The Blessing of Bad Geography in Africa,”
Review of Economics and Statistics, 94(1), 20—36.
N, N., L. W$ (2011): “The Slave Trade and the Origins of Mistrust in
Africa,” American Economic Review, 101(7), 3221—3252.
P$$!, M. (2011): “Economic Discontinuities at Borders: Evidence from Satellite Data
on Lights at Night,” mimeo MIT.
P, L., D. N. W (2010): “Post-1500 Population Flows and the Long-Run
Determinants of Economic Growth and Inequality,” Quarterly Journal of Economics, 125(4),
1627—1682.
R, J. A. (2001): “Social Identity, Inequality, and Conflict,” Economics of Governance,
2(2), 85—99.
W, N. B., J. K. R, L.-E. C (2010): “Representing Ethnic Groups
in Space: A New Dataset,” Journal of Peace Research, 47(4), 491—499.
30
7
7.1
Data Appendix
Cross-Country Data
Income level: Log of per capita GDP at PPP (Chain Index) in 2000. Source: Penn World
Tables, Edition 7. Heston, Summers, and Aten (2011).
Population: Log population in 2000. Source: Penn World Tables, Edition 7. Heston,
Summers, and Aten (2011).
Land Area: Log surface area. Source: Nunn and Puga (2011).
Rule of Law: The rule of law index is "capturing perceptions of the extent to which
agents have confidence in and abide by the rules of society, and in particular the quality of
contract enforcement, property rights, the police, and the courts, as well as the likelihood of
crime and violence." The standardized index which corresponds in 2000 ranges from −2.5 to
+2.5 with higher values indicating better functioning institutions. Source: World Bank Governance Matters Indicators Database (Kaufman, Kraay, and Mastruzzi (2005)). available at:
http://info.worldbank.org/governance/wgi/index.asp
Control of Corruption: The control of corruption index is "capturing perceptions of the
extent to which public power is exercised for private gain, including both petty and grand forms of
corruption, as well as capture of the state by elites and private interests." The standardized index
which corresponds in 2000 ranges from −2.5 to +2.5 with lower values indicating a higher degree
of corruption. Source: World Bank Governance Matters Indicators Database (Kaufmann, Kraay,
and Mastruzzi (2005)). available at: http://info.worldbank.org/governance/wgi/index.asp
Income Inequality. Adjusted Gini coefficient index averaged over the period 1965−1998.
Source: Easterly (2007); based on WIDER.
Ethnic/Linguistic/Religious Fractionalization: Index of ethnic/linguistic/religious
heterogeneity, constructed as one minus the Herfindahl index of the share of the largest ethnic/linguistic/religious groups. It reflects the probability that two randomly selected individuals
follow different ethnolinguistic/religious groups. Source: Alesina et al. (2003).
Ethnic/Linguistic/Religious Segregation: Index ranging from zero to one capturing
ethnic/linguistic/religious segregation (clustering) within countries. If each region is comprised
of a separate group, then the index is equal to 1, and this is the case of full segregation. If every
region has the same fraction of each group as the country as a whole, the index is equal to 0,
this is the case of no segregation. The index is increasing in the square deviation of regional-level
fractions of groups relative to the national average. The index gives higher weight to the deviation
of group composition from the national average in bigger regions than in smaller regions." Source:
31
Alesina and Zhuravskaya (2012).
Ethnolinguistic Polarization 1: Index of ethnolinguistic polarization that achieves a
maximum score when a country is occupied by two groups of the same population. Source:
Montalvo and Reynal-Querol (2005a,b).
Ethnolinguistic Polarization 2: The polarization index accounts for the degree of similarity between linguistic groups using the Ethnologue linguistic tree. Source: Esteban, Mayoral,
and Ray (2012).
Cultural Fragmentation: Index of ethnolinguistic fractionalization that accounts for
the degree of similarity between linguistic groups using the Ethnologue linguistic tree. Source:
Fearon (2003).
Soil quality: Percentage of each country with fertile soil. Source: Nunn and Puga (2012).
Ruggedness: The terrain ruggedness index quantifies topographic heterogeneity. The
index is the average across all grid cells in the country not covered by water. The units for the
terrain ruggedness index correspond to the units used to measure elevation differences. Ruggedness is measured in hundreds of metres of elevation difference for grid points 30 arc-seconds (926
metres on the equator or any meridian) apart. Source: Nunn and Puga (2012).
Tropical: The percentage of the land surface of each country with tropical climate. Source:
Nunn and Puga (2012).
Desert: The percentage of the land surface area of each country covered by sandy desert,
dunes, rocky or lava flows. Source: Nunn and Puga (2012).
Latitude: Absolute latitude is expressed in decimal degrees, for the geographical centroid
of the country. Source: Nunn and Puga (2012).
Gem-Quality Diamond Extraction: Carats of gem-quality diamond extraction between 1958 and 2000, normalized by land area. Source: Nunn and Puga (2012).
Common Law: Indicator variable that identifies countries that have a common law legal
system. Source: La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1999) and Nunn and Puga
(2012).
European Descent: The variable, calculated from version 1.1 of the migration matrix
of Putterman and Weil (2010), estimates the percentage of the year 2000 population in every
country that is descended from people who resided in Europe in 1500. Source: Nunn and Puga
(2012).
Settler Mortality: Log of mortality rates faced by European colonizers in late 19th
century. Source: Acemoglu, Johnson, and Robinson (2001).
State Antiquity: Normalized state antiquity Index in 1950, using a 1% discount rate.
Source: Bockstette, Chanda, and Putterman (2002).
32
Border Straightness Index: The 0 − 1 index reflects how straight -and thus most likely
to be non-organic- national borders are. Source: Alesina, Easterly, and Matuszeski (2011).
Ethnic Partitioning: Percentage of the population of a country that belongs to partitioned ethnic groups. Source: Alesina, Easterly, and Matuszeski (2011).
Regional Fixed Effects: The region constants correspond to: South Asia, East Asia
and Pacific, Latin America and the Caribbean, North America, Western Europe, Eastern Europe
and Central Asia, Middle East and Northern Africa, and Sub-Saharan Africa. The classification
follows World Bank’s World Development Indicators.
Light Density at Night: Light density is calculated averaging light density observations
across pixels that fall within each territory (ethnic/linguistic homeland, Thiessen polygon, and
pixel) and then dividing by population.
Source: Available at http://www.ngdc.noaa.gov/dmsp/global_composites_v2.html.
Water Area: Total area covered by rivers or lakes in square kilometers. Source: Constructed using the "Inland water area features" dataset from Global Mapping International, Colorado Springs, CO, USA. Global Ministry Mapping System.
Elevation: Average elevation in kilometers. Source: National Oceanic and Atmospheric
Administration (NOAA) and U.S. National Geophysical Data Center, TerrainBase, release 1.0
(CD-ROM), Boulder, Colorado. http://www.sage.wisc.edu/atlas/data.php?incdataset=Topography
Land Suitability for Agriculture: Average land quality for cultivation within each
country. The index is the product of two components capturing the climatic and soil suitability
for farming. Source: Michalopoulos (2012); Original Source: Atlas of the Biosphere.
Available at http://www.sage.wisc.edu/iamdata/grid_data_sel.php.
Distance to the Sea Coast: The geodesic distance from the centroid of each country
to the nearest coastline, measured in 1000s of km’s. Source: Global Mapping International,
Colorado Springs, Colorado, USA. Series name: Global Ministry Mapping System. Series issue:
Version 3.0
7.2
Micro-Level Data from Afrobarometer Surveys (3rd Round)
Living Conditions: Respondent’s view of their present living conditions. The question (Q4B)
reads "In general, how would you describe your own present living conditions? ". The answers can
be: (i) very bad, (ii) fairly bad, (iii) neither good nor bad, (iv) fairly good, or (v) very good. For
the district-level analysis responses are averaged across all individuals in each district. Source:
2005 Afrobarometer Surveys.
Education: Respondent’s education/schooling. The question (Q90) reads "What is the
highest level of education you have completed? ". The answers are: 0=No formal schooling, 1=In-
33
formal schooling (including Koranic schooling), 2=Some primary schooling, 3=Primary school
completed, 4=Some secondary school/ High school, 5=Secondary school completed/High school,
6=Post-secondary qualifications, other than university e.g. a diploma or degree from a technical/polytechnic/college, 7=Some university, 8=University completed, 9=Post-graduate. For the
district-level analysis responses are averaged across all individuals in each district. Source: 2005
Afrobarometer Surveys.
Access to piped water: Response to the question (Q116e) on "whether in the enumeration area there is a piped water system that most houses could access". For the district-level
analysis responses are averaged within a district. Question was filled in conjunction with field
supervisor. For the district-level analysis responses are averaged across all individuals in each
district. Source: 2005 Afrobarometer Surveys
Access to sewage system: Response to the question (Q116f) on "whether in the enumeration area there is a sewage system that most houses could access". For the district-level
analysis responses are averaged within a district. Question was filled in conjunction with field
supervisor. For the district-level analysis responses are averaged across all individuals in each
district. Source: 2005 Afrobarometer Surveys.
Access to an electricity grid: Response to the question (Q116d) on "whether in the
enumeration area there is an electricity grid that most houses could access". For the district-level
analysis responses are averaged within a district. Question was filled in conjunction with field
supervisor. For the district-level analysis responses are averaged across all individuals in each
district. Source: 2005 Afrobarometer Surveys
Support for Democracy: Ordered (0, 1, 2) variable capturing support for democracy,
based on the following question (Q37). "Which of these three statements is closest to your own
opinion? A: Democracy is preferable to any other kind of government. B: In some circumstances,
a non-democratic government can be preferable. C: For someone like me, it doesn’t matter what
kind of government we have." We assign a value of 2 for individuals who reply A, a value of
1 for individuals who choose B, and a 0 for individuals who choose C. For the district-level
analysis responses are averaged across all individuals in each district. Source: 2005 Afrobarometer
Surveys.
Satisfaction with Democracy: Ordered (0, 1, 2, 3, 4) variable reflecting satisfaction with
democracy. The question (Q47) reads. "Overall, how satisfied are you with the way democracy
works in [Ghana/Kenya/etc.] are you?" The replies are 0=My country is not a democracy, 1=Not
at All Satisfied, 2=Not Very Satisfied, 3=Fairly Satisfied, 4=Very Satisfied. For the district-level
analysis responses are averaged across all individuals in each district. Source: 2005 Afrobarometer
Surveys.
34
Constitution: Ordered variable (1, 2, 3, 4, 5) reflecting people’s beliefs (satisfaction)
about whether the constitution expresses the values and hopes of people. The question (Q52A)
reads: Constitution expresses values and hopes. Do you disagree or agree with the following
statement: Our constitution expresses the values and hopes of the [Ghanaian/Kenyan/etc.] people. : 1=Strongly Disagree, 2=Disagree, 3=Neither Agree Nor Disagree, 4=Agree, 5=Strongly
Agree. For the district-level analysis responses are averaged across all individuals in each district.
Source: 2005 Afrobarometer Surveys.
Ethnic-level Economic Conditions (power): Ethnic-level average of individual responses to the following question (Q80A) reflecting economic conditions of each group in the
country. "Think about the condition of _ _ _ _ _ _ _ _ _ _ _ _ [respondent’s identity group]. Are their economic conditions worse, the same as, or better than other groups in
this country? " 1=Much Better, 2=Better, 3=Same, 4=Worse, 5=Much Worse. Source: 2005
Afrobarometer Surveys.
Ethnic-level Political Conditions (power): Ethnic-level average of individual responses to the following question (Q80B) reflecting political conditions of each group in the
country. "Think about the condition of _ _ _ _ _ _ _ _ _ _ _ _ [respondent’s identity group]. Are their political conditions worse, the same as, or better than other groups in
this country? " 1=Much Better, 2=Better, 3=Same, 4=Worse, 5=Much Worse. Source: 2005
Afrobarometer Surveys.
Ethnic-level Unfairness (Discrimination): Ethnic-level average of individual responses
to the following question (Q81) reflecting unfairness (discrimination) of each group by the government of each country."How often are _ _ _ _ _ _ _ _ _ _ _s [respondent’s identity group]
treated unfairly by the government?" 0=Never, 1=Sometimes, 2=Often, 3=Always. Source: 2005
Afrobarometer Surveys.
35
Ethnic Homelands in Afghanistan
Afghanistan Arabs
Afghans
Arabs of Middle Asia
Baloch
Brahui
Burushaskis
Firoz-Kohis
Hazara-Berberi
Hazara-Deh-i-Zainat
Ishkashimis
Jamshidis
Kazakhs
Kho
Kirghis
Mongols
Nuristanis
Ormuri
Pamir Tajiks
Parachi
Pashai
Uzbeks
29
70
71
-0
.2
3
14
.0
1
.0
0
-0
15
-0
0.
01
09
0.
00
10
0.
00
0.
00
Turkmens
.0
0
.0
0
02
Ü
Tirahi
-0
Ü
Teymurs
-0
Tajiks
03
Taimanis
00
Shugnanis
0.
00
Russians
0.
00
Roshanls
08
Lights per Capita in Ethnic
Homelands in Afganistan
Atlas Narodov Mira
Persians
Figure 1b
Lights per Capita across Ethnic Homelands
Yazghulems
Overlapping Languages
Figure 1a
Ethnic Homelands in Afghanistan
Lights per Capita across
Thiessen Polygons
in Afganistan
High : 0.0078
Low : 0
09
3
02
1
-0
.0
01
6
.0
-0
22
0.
00
00
7
.0
-0
17
0.
00
00
3
.0
08
0.
00
04
-0
.0
.0
-0
0.
00
-0
00
02
0.
00
0.
00
Ü
00
1
Lights per Capita across
Virtual Countries in Afganistan
Figure 2a: Lights across 2.5 by 2.5 dd Boxes
36
ÜÜ
Lights
Capita
across
Lights
perper
Capita
across
Thiessen
Polygons
Thiessen
Polygons
in in
Afganistan
Afganistan
High
: 0.0078
High
: 0.0078
Low : 0
Low : 0
Figure 2b: Lights across Thiessen Polygons
Figure 3e
37
7
Figure 3f
0.2
79
0.1
96
0.1
45
21
1
01
3
02
8
48
5
0.1
17
99
50
4
0.0
.1
05
15
8
79
01
2
.4
-0
4
0
21
45
48
96
02
7
.2
-0
.1
-0
-0
-0
.1
17
99
50
3
71
04
9
59
7
5
36
28
0.9
97
15
0.8
3
38
51
1
95
03
0.8
0.7
8
40
24
0.7
2
86
28
0.6
3
38
97
0.6
8
6
05
93
0.6
42
97
0.5
8
7
91
24
0.4
50
67
0.4
6
41
1
01
32
0.4
0.3
0
24
87
0.3
3
2
83
13
0.2
25
32
0.2
9
0
78
54
0.1
33
87
0.1
9
60
84
0.0
9
3
1
5
2
9
2
8
5
0
66
57
16
87
87
70
00
42
79
.9
.8
.7
.7
.6
.6
.5
.5
.5
58
86
66
3
.9
-0
54
36
2
.9
-0
97
1
.8
-0
12
30
38
5
.8
-0
95
0
47
40
2
.7
-0
81
86
2
.7
-0
.6
-0
72
38
9
37
05
9
.6
-0
.6
-0
75
42
9
47
91
2
.5
-0
.4
-0
5
76
32
10
50
6
01
59
4
.4
.4
-0
-0
.3
-0
69
24
8
32
83
1
.3
-0
.2
-0
21
25
3
.2
-0
48
78
5
.1
-0
69
33
8
.1
-0
8
8
2
0
4
1
8
1
7
48
67
66
57
16
87
87
70
00
42
4
30
79
60
8
75
69
78
22
83
38
86
41
3
6
1
1
9
2
9
5
11
7
11
6
50
00
.4
.4
.0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
24
31
25
62
64
09
82
38
76
36
01
59
29
69
29
80
25
.3
.3
.3
.2
.2
.1
.1
.0
.0
-0
-0
-0
-0
-0
-0
-0
-0
-0
0.7
0.7
0.6
0.6
0.6
0.6
0.6
0.5
0.5
0.5
0.4
85
17
84
69
53
36
08
72
38
09
2
94
01
90
21
08
75
93
85
80
78
1
7
5
8
2
6
4
4
6
5
6
11
3
48
26
78
0.4
5
0
8
3
33
69
17
36
94
67
06
69
15
72
0.3
0.3
0.2
0.2
0.1
1
4
63
0
6
6
0
2
0
6
5
5
0
1
4
0
1
9
5
7
75
47
0.1
12
0.0
9
9
00
14
33
02
06
73
40
65
70
47
18
46
22
57
82
83
99
21
00
20
92
56
29
95
55
91
65
46
06
36
01
71
83
21
63
0.0
0.9
0.8
0.8
0.8
0.7
0.7
0.6
0.6
0.6
0.6
0.5
0.4
0.3
0.2
0.2
0.1
0.1
13
1
0
69
65
00
81
28
00
0.0
0.0
0.0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
.9
.7
.7
.6
.6
.6
.6
.6
.5
.5
59
85
17
84
69
53
36
08
72
38
1
4
9
69
94
01
90
21
08
75
93
85
80
5
2
0
6
4
7
1
5
3
3
5
4
48
78
78
09
.4
.5
-0
-0
2
7
11
2
33
69
16
36
94
3
0
0
5
5
9
1
9
5
4
4
9
0
3
9
0
8
75
26
67
06
69
15
72
.4
.3
.3
.2
.2
6
4
63
.1
12
.1
-0
-0
-0
-0
-0
-0
-0
64
14
33
01
06
72
40
65
70
46
18
46
21
57
82
83
47
75
20
92
56
29
95
55
91
65
46
06
36
01
71
83
21
99
8
8
0
21
69
65
63
.0
.9
.9
.8
.8
.8
.7
.7
.6
.6
.6
.6
.5
.4
.3
.2
.2
.1
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
-0
81
28
.1
13
.0
.0
-0
-0
-0
Ü
.0
-0
.0
-0
Ü
05
0
93
4
49
93
3
.0
35
32
9
.0
-0
-0
21
5
21
.0
-0
.0
11
34
6
-0
69
78
22
83
38
86
41
00
50
31
11
8
11
6
76
00
00
0.0
0.8
0.7
0.7
0.6
0.6
0.5
0.5
0.5
0.4
0.4
0.3
4
7
2
2
0
3
0
6
0
24
31
25
62
65
09
83
38
00
36
01
59
29
69
29
80
25
00
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0.0
0.0
Ü
0.0
71
0.0
49
33
0
21
6
11
34
7
51
1
90
3
.0
02
- -0
9
54
5
26
81
2
52
- -0
.0
18
90
2
0.0
35
.0
.0
32
- -0
- -0
.0
71
- -0
56
8
25
3
.0
93
- -0
.1
26
- -0
1
6
9
7
22
4
.2
41
- -0
39
27
21
65
40
2
32
27
10
62
Figure 3c
0.0
21
.4
.3
.2
.1
3
6
7
8
2
0
5
61
01
54
80
75
16
58
1
3
43
.1
19
97
82
68
53
46
31
.4
00
-0
-0
-0
-0
-0
- -0
7
0
8
4
.0
.0
.0
.0
.0
.0
57
14
00
.0
1
0
9
5
0
0
0
Figure 3a
0.0
44
51
0
.0
02
-0
18
-0
.0
26
8
81
1
56
7
25
2
22
3
40
1
32
5
-0
.0
27
22
65
96
8
71
52
.0
-0
.0
-0
93
-0
.0
26
-0
.1
41
-0
.2
00
06
27
10
62
61
-0
-0
-0
-0
-0
-0
-0
92
67
55
91
95
08
66
Ethnic Inequality Partialling Out Spatial Inequality Based on Thiessen Polygons: Ethnologue
-0
.4
-0
.7
0.3
0.2
0.1
0.1
19
8
9
3
1
6
2
4
01
54
80
75
16
58
43
57
97
82
68
53
46
31
14
00
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
17
53
77
99
63
79
79
.0
.0
.0
.0
.0
.1
.2
.4
-0
- -0
- -0
- -0
- -0
- -0
- -0
- -0
0
9
8
4
9
9
9
4
92
66
55
91
94
07
17
53
77
99
63
79
.0
.0
.0
.0
.1
.2
-0
-0
-0
-0
-0
-0
65
51
18
79
.7
.4
-0
-0
Ethnic Inequality in 2000 Based on the Ethnologue Mapping
Ethnic Inequality in 2000 Based on the Atlas Narodov Mira Mapping
Ü
Figure 3b
Ü
Spatial Inequality Based on 2.5 by 2.5 Decimal Degree Boxes
Spatial Inequality Based on Thiessen Polygons: Centroids are those of Ethnologue's Groups
Figure 3d
Ethnic Inequality Partialling Out Spatial Inequality Based on Thiessen Polygons: Atlas Narodov
Ü
Unconditional Relationship
1
Ethnic Inequality and Overall Spatial Inequality, GREG
Unconditional Relationship
1
Ethnic Inequality and Overall Spatial Inequality, GREG
SDN
SDN
AFG
AFG
MNG
.2
.4
.6
Overall Spatial Inequality (Index 1)
.8
.8
LAO
ZAR
CAF TCD
NGA
SLE
SLB
COG
BWA
GAB
ZMB DZA MLI GNQ
AGO
CMRIDN
GUY TGO
BRA
NPL
ETH NER
BFA GIN
SUR
UGA
GNB
ECU MRT
BEN
TKM CHL KEN
VEN
BRN
PRY MOZ
NIC
RWA TZA
SOM
GHATJK
THA
PAK MWI
COL
ERI
SENZWE
LBY
PHLVNM KAZ
AUS
GMB
LSO
BDI
NOR
NAM KHM
BGR
GTM
SWE
CAN URY
BTN
SWZ
MYS
JOR
SLV CIV USA
CRI
LVA IRN
BGD
ALB QAT
ISR DJI
HRV
KWT
HTI
MEX
MKD FJI
PNG
UZB
DOM
FRA
SAU
SVN ROMGRC
BIH
ITA
DEU GBR
KGZZAF JPN
BLR
MAR
UKR
CHE
OMN
BEL
AUT
LKA SVK
LTU
TUR
DNK
ESP
EST
MDA CZE
PRT
NLDIRL
POL
TTO
CYP LUX HUN
LBN
SGP
ARE NZL
VUT
WSM
GRD
MUS
TON
ATG
CPV
KNA
JAM
MLT
VCT
LCA
ISLDMA
BHRBHS
CUB
KOR COM
MDG
SYC
.6
0
1
.2
.4
.6
Overall Spatial Inequality (Index 2)
Figure 4a
Unconditional Relationship
1
1
Ethnic Fragmentation and Ethnic Inequality, GREG
Unconditional Relationship
MDA
LUX
EST
BEL
CHE
MAR
UKR
OMN
ESP
LKA
NZL
SGP
CZE TUR
LTU
VCT
GRD
DOM
UZB
HRV
ISR
BLR
ROM
PNG
SVK
TON
ISL
MLT
VUT
KOR
COM
HUN
LBNPOL
IRL
NLD
CYP
AUT
DNK
SVN
SAU
DEUGRC
GBRITA
FRA
BDI
LSO
PHL
CRI URY VNM
ALB
KHM
SLV
BGD
.8
IND
BWA
IRQ
GNQ
DZA
RWA
HTI
PRT
TKM
ZWE
ARG
EGY
UKRMAR
OMN
DOM
LKA
SGP NZL
IRL
HTI
DNK
PRT
VUT
JPN
TUN
JPN
BEL
CHE
MDA
LUX
ESP
CZE TUR
BDI
LSO
SLB
KGZ
KWT
TTO
EST
RUS
AZE
CHL HND
PRY
CHN
FIN
ARM
AUS
SWE
SWZ
NOR
MNG
ARE
UZB
ROM
PNG
SVK
URY
SVN
DEUGRCSAU
GBR
ITA
AUT
FRA
HUN
LBN
POL
NLD
IND BWA
TKM
IRQ
GNQ
DZA
RWA
BGR
ZWE
HRV
ISR
LTUBLR
CRI
ALB
ARG
VNM
PHL
SLV KHM
AZE
HND
CHL
PRY
CHN
FIN
SWZ NOR SWE
0
.2
.4
.6
Gini Ethnic Inequality in 2000; GREG
.8
0
1
MNG
EGY
ARM
SLB
TUN
.2
.4
.6
.8
Gini Ethnic Inequality in 2000 (excl. capital cities); GREG
Figure 5a
Unconditional Relationship
70
Income Inequality and Ethnic Inequality, ETHNOLOGUE
Unconditional Relationship
70
Income Inequality and Ethnic Inequality, GREG
GAB
CAF
SLE
TUR
MEX
DOM
BHS
TTO
IRL
MDA
PRT
EST LKA
LTU
40
JAM
MUS
KGZ
BIH
SGP
NLD
DNK
SEN NIC
GTM
COL
GMB
PANMLI
BFA
ARM ZMB
GUY
GIN
CHL
BWA
MYS
PHL
NPL TUN NGA
AZE
THA VENECU
ARG
CRI
MRT
KHM
ETH
NER
JOR URY
CIV
TKM
RUS
UGA DZA
USA
AUS
VNM
NOR
GRC
ITA
CHE
FRA
MKD
GBR
DEU
ISRLVA
BGD
ESP
UKR
JPN ROM
BEL
HUNPOL
BLR SVN
CZE
AUT
LUX
SVK
NZL
30
FJI
SLV
GEO
BDI
GHA
SWE KAZPAK
CAN
RWA
IDN
IND
FIN CHN
BOL
PER
SDN
TCD
LAO
EGY
SUR
BGR
60
ZAF
SWZ
ARM
BHS
MDA
KEN
PRY
SLV
MEX
TUN
IRL
TTOKGZ
BWA
PRT
LKA
LTU
BIH
SGP
GRC
FRA
MKD
DNK
GBR
NLD
NZL
BDI
DEU
ISR
KAZ
LVA
KOR
ROM UKRJPN
BEL
RWA
POL
BLR SVN
CZEHUN
AUT
LUX
BGRSVK
JOR
MRTCRI
CIV
ZWE
BRA
CHL
FJI MYS
AZE
THA ECU
ARG
USA
BGD
CAN
PNG
MLIPAN
NICSEN
GTM
MDG
TUR ZMB
GMB
ITA
CHE
NOR
ESP
CAF
TZA
GEO
DOM
JAM
EST
MUS
URY
IRQ
LSO
HND
LBN
50
50
PNG
40
LBN
IRQ
ZWE PRYKEN
LSO
BRA
HND
TZA
Gini Coefficient - Income Inequality
SWZ
SLE
30
60
ZAF
MDG
1
Figure 5b
GAB
KOR
SDN
RUS
AUS
BGD
AFG
0
SYC
DMA
KNA
LCA
ATG
WSM
BGR
ZAF
QAT
SDN
PER
CIV
GMB LBY
DJI
.6
CUB
AFG
BOL
.4
BIH
LBR
ZAR
CMR COG
TCD
NGA
CAFSLE
SOM
GNB
BEN AGO
ZMBGAB
BOL
GIN
BFA
TZA
IDN
SUR
CAN
PAK
TGO BLZETH
SEN
MOZ
MLI
MWI
GHA NPL
IRN
PER
ERI ECU
NER
THA
NAM
BIH
GUY
KAZ
MRT
BTN
COL
JOR LVA
MYS
FJI
MEX
BRN
BRA PAN
SYR
LAO
TJK
MKD USA GTM
VEN
GEO
NIC
KEN
GNB
BEN
Ethnic Fragmentation
.6
ARE
TTO
SOM
.2
.8
KGZ
.4
Ethnic Fragmentation
ZAF
BHR
MUS
BHS
CPV
JAM
UGA
LBR
ZAR
CMRCOG NGA
TCD
CAF
SLE
DJI
LBY
GMB
AGO
ZMB
GAB
QAT
GIN
BFA
TZA
IDN
SUR
ETH
CAN
PAK
TGO
BLZ
SEN MOZ
MLI
MWI
GHA
NPL
KWT IRN
ECU
ERI
NER
NAM THA
MRT GUY
BTN KAZCOL
JOR
LVA MYS
PAN
FJI
MEX
BRN SYR
BRA
LAO
GTM
TJKVEN
MKD
GEO
USA
NIC
KEN
CIV
VNM
GHA
IND
SWE PAKCHN
COL
BFA
GINGUY
BOL
PHL
KHM
TKM
RUS
DZAUGA
NPLPERNGA
VEN
NER ETH
SDN
AUS
IDN TCD
FIN
EGY LAO
SUR
MNG
MNG
20
20
Gini Coefficiet - Income Inequality
1
Ethnic Fragmentation and Ethnic Inequality, GREG
MDG
.2
.8
Figure 4b
UGA
0
RUS
IRQ
CHN
IND
ARM
BLZ
PAN
AZE ARG
SYR
FIN
HND
0
.6
.4
.2
0
LBR
PER
BOL
GEO
TUN
.4
Gini Ethnic Inequality in 2000; GREG
LAO
ZAR
TCD
CAF
NGA
SLE
SLB RUS
COG
BWA
GAB
ZMB
AGO
MLI
IRQ CMR
CHN
DZA
GNQ
GUY
TGO
IDN
NPL
ETH
INDBRA
GIN
BFA
NER
AZE
SYRBLZ
FIN
ARG
HND
SUR
UGA BEN
MRT
GNB
ECU
KEN
TKM
VEN
CHL
BRN
MOZ
PRY
NIC
TJK
RWA
TZA
SOM
GHA
THA
MWI PAK
COL ERI
SEN
ZWE
PHL
KAZ LBY
AUS
VNM
KHM LSO
GMBNOR
BDI
NAM
BGR
GTM
BTN
CAN
URY SWE
SWZ
MYS
USA
CIV
JOR
SLV
CRI
LVA
BGD
ALB
QAT
IRN
DJI
ISR
HRV
KWT
HTI
MEX
MKD
FJI
PNG
UZB
DOM
FRA
SAU
SVNROM
GRC
BIH
ITA
DEU
KGZ
ZAF JPN
BLRGBR
MAR CHE UKR
OMN
BEL
AUT
LKA
SVK
LTU
TUR
DNK
ESP
EST
MDAPRT
CZE
NLD
POL
TTOIRL
LUX
CYP
LBN
HUN
SGP
ARE
NZL
VUT
WSM
GRD
BHR
DMA
ATG
KNA
JAM
MLT
VCT
TON
MUS
LCA
KOR
CUBCPVISL
COM
SYC
BHS
MDG
GEO
.2
.8
BOL
EGY
PER
TUN
ARMPAN
0
Gini Ethnic Inequality in 2000; GREG
MNG
LBR
EGY
0
.2
.4
.6
Gini Ethnic Inequality in 2000; GREG
.8
0
1
Figure 6a
.2
.4
.6
Gini Ethnic Inequality in 2000; ETHNOLOGUE
Figure 6b
38
.8
1
Ethnic Inequality and Comparative Development
Ethnic Inequality and Comparative Development
Unconditional Relationship
12
12
Unconditional Relationship
PER
EGY
BOL
MNG
SDN
LBR
AFG
10
8
Ln(GDP per Capita) in 2000
LUX
BRN
NOR
USA
KWT
SGP NLD
ARE
CHE
ISL
CAN AUS
BEL GBR
IRL DNKAUT
JPN
DEU
SWE
ITA FRA
FIN
ESP
BHS NZL
ISR
GRC
BHR
MLT
PRT
SVN
KOR
SYC
OMN
CYPCZE
SAU
LBY
TTO
ATG
HUN
SVK
GRD
POL
GAB
LCA
MEX
EST
KNA
HRV CRIMYS
CHL ARG
LBN
LTU
JAM
URY
TUR
CUB
VEN BLZ
LVA
IRN
BRA BWA RUS
MUS
DOM
SURPAN
TON
BGR
BLR
ROM MKD
ZAF BIH
GNQ
THA
WSM
VUT
SLV GTM KAZCOL
DMA
ECU DZA
TUN
TKM
VCT
IRQ
FJI
NAM
UKR
ALBJORSWZ
SYRGUY
PRY
HND
GEO
AZEIDN
CHN
LKA MAR
BTN
ARMAGO
CPV
PHL
COG
PNG
NIC
DJI
IND
CMR
VNM PAK
KGZ
MDA
HTI
CIV
UZB
SLB LAO
SEN
MRT
BEN
KEN
NGA
LSO
KHM
NPL
TJK
BGD
COM
GHA
TGO
ZMB
UGA
MDG
GMB ERI
BFA
GIN
MLI
TCD
TZA
RWA
CAF
MWI
SLE
NER
SOM
ETH
GNB
BDIZWE MOZ
QAT
LUX
KWTSGP
ARE
NLD
ISL
DNK
BEL
IRL
DEU
JPN AUT
GBR
FRA
BHS
NZL
ISRGRC
BHR
MLT
KOR
SYC SVN
SAU CZE
CYP PRT
TTO
ATG
HUN
SVK
POL
GRD
LCA
EST
KNA
HRV
LBN
LTU
JAM
URY
CUB
LVA
MUS
DOM
TON
BGR
MKD
BLR
ROM
ZAF
BIH
WSM
KAZ
DMASLV
TUN
VCT
UKR
SWZ
LKA
MAR
CPVARM
DJI
KGZ
HTIMDA
UZB
RWA
6
10
8
6
Ln(GDP per Capita) in 2000
QAT
NOR
CHE
ITA
ESP
AUS
FIN
OMN
LBY
BLZ BWA
MEX
CRI
GAB
MYS
RUS VEN
IRN
BRA
SURPAN
GNQ
COL VUT
GTM
PER
DZA
ECU
TKM
FJI
EGY
SYR BOL
PRY
GUY
IDN
AZE
CHN
BTN
PHL
AGO
COG
PNG
NIC MNG
IND
PAK VNM
CMRSDN
LAOSLB
SEN
KEN BEN
NPL NGA
TJK KHM
GHA
TGO
ZMB
MDG ERI
GMB
GIN BFA
MLI UGA
TCD
TZA
CAF
SLE
LBR
GNB NER ETH
MOZ SOM
AFG
ZWE
CHL
ARG
TUR
THA
HND
IRQ
JOR ALB
GEO
COM
CIV
MRT
LSO BGD
MWI
BDI
BRN
USA
SWE
CAN
NAM
ZAR
4
4
ZAR
0
.2
.4
.6
Gini Ethnic Inequality in 2000; GREG
.8
1
0
.2
Figure 7a
.8
Inequality in Geography and Contemporary Ethnic Inequality
Unconditional Relationship
Unconditional Relationship
-4
-2
0
2
Inequality in Geographic Endowments across Ethnic Homelands
1st Principal Component
.8
.6
BWA
MWI ESP
BLZ CHE
.4
Gini Coefficient - Ethnic Inequality in 2000
ETHNOLOGUE
CHL
WSM
RWA
SWZ
KOR
MUS
CUB
TON
URY
ATG
BHS
CPV
HRV
KNA
QAT
SYC
BLR
EST
JAM
MLT
VCT
LCA
LUX
NZL
BDI
HTI
ISL
0
.8
.6
.4
.2
0
ARE
WSM
COM
MDG
GRD
KOR
MUS
BHR
CUB
DMA
TON
ATG
BHS
CPV
KNA
SYC
JAM
MLT
VCT
LCA
ISL
MNG
PER
EGY
BOL
LAO
ZAR
GEO
TCD
CAF
NGA
RUS
SLE
SLB
COG GAB
BWA
TUN
ZMB
AGO
MLI
IRQ
CHN
DZA
CMRGNQ
GUY
TGO PAN
IDN
NPL
BRA
ETH
IND
ARM
GIN
BFA
BLZ HND
NER AZE
UGA SUR
MRT SYR FIN ARG
ECU GNB
BEN
KEN
TKM
MOZ VEN
PRY BRN
NIC
TJK
RWA
TZA
GHA
THASOM
PAK
MWI
COLERI
ZWE SEN
LBYKAZ
PHL
AUS
VNM
KHM
GMB
BDI
LSO
NAM
BGR NOR
GTM
SWE
CAN
BTN
SWZ URY
MYS
USA
CIV CRI
JOR
SLV
LVA BGD
ALB
QAT
IRN
DJI
ISR
HRV
HTIFJI KWT
MEX
MKD
PNG
UZB
DOM
SAUFRA
ROM
SVN
GRC
BIH
ITA
DEU
JPN
ZAF
BLR GBR KGZ
UKR
CHE MAR
OMN
BEL
AUT
LKA
LTUSVK
TUR
DNK
ESP
EST
MDA
PRT
IRL LUX
NLD
POL
TTOCZE
CYP
LBN
HUN
SGP
NZL
VUT
TCD
SDN
PNG
BRA
ETH
ZAR
COG
CMR
IDN SLB
LAO
VUT
AGO
NGA
ZWE
NER
AFG
AUS
CAF
PER
BFA LBR
COL
GAB
EGY
GNB
NPL
GUY VEN
BTN
OMN
TGOGIN
BOL
TZA
SLE
RUS
KHM
UGA
PHL
SYR GNQ
TKM
PAN
FIN TJK
DZA
MNG
BRN SUR GTM
MLI MYSSOM
SEN
NIC
VNM FJI IRN CHN
MOZ
BEN
ECU
ZMB
ERI LBY AZE
KEN
PAK
GHA
PRY
THA
TUR
MDG
USA
CHL
SWE
GMB
NAM ARG
IND
CRI
ALB
CIV
MRT
BGD
GEO
MEX
IRQ
CAN
LSOJOR
1
SDN
.2
1
Inequality in Geography and Contemporary Ethnic Inequality
LBR
1
Figure 7b
AFG
Gini Coefficient - Ethnic Inequality in 2000
GREG
.4
.6
Gini Ethnic Inequality in 2000; ETHNOLOGUE
-4
4
Figure 8a
HNDNOR
ITA
PRT
BIH DNK TUN KAZGRC
ZAF
SGP
ISR
CYP
UZBAUT
KGZ
KWT
DEU
GBR
UKR
DJI
SVK
MKD
DOM
LKA
HUN
FRA
LTU
BGR ROM
NLD
BEL
CZE
IRL
MAR
SLV
BHR
POL
SAU SVN
MDA LVA
ARM
ARE
LBN
-2
0
2
Inequality in Geographic Endowments across Ethnic Homelands
1st Principal Component
4
Figure 8b
Group Living Conditions and Group Discrimination by the Government
Conditional on Country Fixed Effects and Number of Respondents for Each Group
Conditional on Country Fixed Effects and Number of Respondents for Each Group
-1
-1
-.5
-.5
0
0
.5
.5
Discrimination of the Group by the Government
Political Influence of the Ethnicity Relative to other Groups
1
1
Group Living Conditions and Relative Political Influence of the Group
-1
-.5
0
.5
Mean Individual Living Conditions within each Group
-1
1
Figure 9a
-.5
0
.5
Mean Individual Living Conditions within each Group
Figure 9b
39
1
Conditional on Country Fixed Effects and Number of Respondents for Each Group
-.5
0
.5
Economic Condition of the Ethnicity Relative to other Groups
1
.5
0
-.5
-1
-.5
0
.5
1
Mean Individual Living Conditions within each Group
-1
-1
-.5
0
.5
Economic Conditions of the Ethnicity Relative to Other Groups
Figure 10a
Figure 10b
Group Economic Conditions and Group Discrimination by the Government
-.5
0
.5
Discrimination of the Group by the Government
1
Conditional on Country Fixed Effects and Number of Respondents for Each Group
-1
-1
1
Economic and Political Power at the Ethnicity Level
Conditional on Country Fixed Effects and Number of Respondents for Each Group
Political Influence of the Ethnicity Relative to other Groups
Group Living Conditions and Relative Economic Status of the Group
-1
-.5
0
.5
Economic Conditions of the Ethnicity Relative to Other Groups
Figure10c
40
1
1
Table 1A: Summary Statistics - Cross Country Inequality Measures
Obs.
mean st. dev.
p25
median
p75
min
max
Number of Ethnicities (GREG)
Ethnic Gini in 2009 (GREG), All Groups
Ethnic Gini in 2000 (GREG), All Groups
Ethnic Gini in 1992 (GREG), All Groups
173
173
173
173
11.52
0.42
0.42
0.48
14.17
0.26
0.26
0.29
1.00
0.00
0.00
0.00
3.00
0.19
0.19
0.21
8.00
0.47
0.47
0.56
13.00
0.63
0.65
0.72
94.00
0.96
0.96
0.97
Number of Languages (ETHNOLOGUE)
Ethnic Gini in 2009 (ETHNOLOGUE), All Groups
Ethnic Gini in 2000 (ETHNOLOGUE), All Groups
Ethnic Gini in 1992 (ETHNOLOGUE), All Groups
173
173
173
173
41.91
0.45
0.46
0.50
99.78
0.33
0.34
0.35
1.00
0.00
0.00
0.00
3.00
0.13
0.11
0.15
9.00
0.47
0.51
0.55
36.00
0.77
0.77
0.83
791.00
0.97
0.98
0.99
Number of Pixels
Spatial Gini in 2009, Pixels
Spatial Gini in 2000, Pixels
Spatial Gini in 1992, Pixels
173
173
173
173
24.28
0.40
0.40
0.45
63.84
0.26
0.26
0.27
1.00
0.00
0.00
0.00
4.00
0.17
0.17
0.21
8.00
0.40
0.40
0.47
22.00
0.60
0.60
0.68
637.00
0.98
0.98
0.95
Number of Thiessen Polygons
Spatial Gini in 2009, Thiessen Polygons
Spatial Gini in 2000, Thiessen Polygons
Spatial Gini in 1992, Thiessen Polygons
173
173
173
173
50.79
0.47
0.48
0.52
98.05
0.29
0.29
0.31
1.00
0.00
0.00
0.00
7.00
0.21
0.23
0.23
17.00
0.48
0.46
0.52
54.00
0.71
0.73
0.81
698.00
0.97
0.97
0.99
Table 1B: Summary Statistics - Afrobarometer Sample - District Level
Theil Index - Overall Inequality
Theil Index - Between-Group Inequality
Theil Index - Within-Group Inequality
Living Conditions Index
Education
Access to Sewage System
Access to Clean Piped Water
Access to Electricity Grid
Ethnic Fractionalization
Number of Ethnic Groups per District
Number of Respondents per District
Support for Democracy
Satisfaction with Democracy
Constitution Expresses People's Hopes-Values
Obs.
mean st. dev.
p25
median
p75
min
max
1298
1298
1298
1298
1298
1277
1283
1289
1298
1298
1298
1284
1286
1290
0.05
0.04
0.01
2.71
4.01
0.24
0.47
0.54
0.71
2.74
16.29
1.57
2.60
3.52
0.00
0.00
0.00
2.25
3.00
0.00
0.00
0.00
0.50
1.00
7.00
1.38
2.13
3.15
0.04
0.03
0.00
2.75
4.00
0.00
0.44
0.60
0.76
2.00
8.00
1.67
2.67
3.59
0.08
0.07
0.02
3.17
5.00
0.50
1.00
1.00
1.00
3.00
16.00
1.86
3.04
3.98
0.00
0.00
0.00
1.00
1.00
0.00
0.00
0.00
0.11
1.00
1.00
0.00
0.00
1.00
0.31
0.21
0.31
4.88
8.75
1.00
1.00
1.00
1.00
18.00
357.00
2.00
4.00
5.00
0.05
0.04
0.02
0.67
1.42
0.39
0.45
0.45
0.27
2.24
23.16
0.39
0.67
0.61
Table 1C: Summary Statistics - Afrobarometer Sample - Individual Level
Obs.
Theil Index - Overall Inequality
Theil Index - Between-Group Inequality
Theil Index - Within-Group Inequality
Living Conditions Index
Education
Access to Sewage System
Access to Clean Piped Water
Access to Electricity Grid
Ethnic Fractionalization
Number of Ethnic Groups per District
Number of Respondents per District
Support for Democracy
Satisfaction with Democracy
Constitution Expresses People's Hopes-Values
mean st. dev.
21138 0.07
21138 0.06
21138 0.01
21138 2.63
21138 4.09
20487 0.24
20818 0.51
20878 0.55
21138 0.60
21138 4.85
21138 49.19
18199 1.58
18024 2.56
17733 3.51
0.05
0.05
0.02
1.20
2.02
0.43
0.50
0.50
0.35
3.93
63.35
0.73
1.06
1.18
p25
median
p75
min
max
0.03
0.02
0.00
2.00
3.00
0.00
0.00
0.00
0.25
2.00
12.00
1.00
2.00
3.00
0.07
0.06
0.00
3.00
4.00
0.00
1.00
1.00
0.69
4.00
24.00
2.00
3.00
4.00
0.11
0.09
0.02
4.00
5.00
0.00
1.00
1.00
0.94
7.00
63.00
2.00
3.00
4.00
0.00
0.00
0.00
1.00
1.00
0.00
0.00
0.00
0.00
1.00
1.00
0.00
0.00
1.00
0.31
0.21
0.31
5.00
10.00
1.00
1.00
1.00
1.00
18.00
357.00
2.00
4.00
5.00
Panel A reports summary statistics for the main ethnic inequality and overall spatial inequality measures employed in the cross-country
analysis. Section 3 gives details on the construction of these measures.
Panel B reports summary statistics for all measures, employed in the cross-district analysis within African countries (Afrobarometer
sample). Panel C reports summary statistics for all measures, employed in the cross-individual analysis within African countries
(Afrobarometer sample). The Data Appendix gives detailed variable definitions and sources.
Table 2a - Baseline Estimates: Ethnic Inequality and Economic Development (in 2000), Atlas Naradov Mira
(1)
Ethnic Inequality
[Gini Coeff., GREG]
(2)
-1.4707***
(0.2504)
-5.87
Spatial Inequality 1
(4)
-1.4003***
(0.3633)
-3.85
-1.1508***
(0.2786)
-4.13
[Gini Coeff., Pixels]
(3)
(5)
(6)
-1.4985***
(0.4018)
-3.73
(7)
-0.1186
(0.3809)
-0.31
-1.1612***
(0.2559)
-4.54
[Gini Coeff., Thiessen Polyg]
(9)
0.0373
(0.3961)
0.09
Log Number of Languages
[GREG]
0.0880
(0.4084)
0.22
-0.3037*** -0.0887
(0.0605) (0.0930)
-5.02
-0.95
0.661
173
Yes
0.626
173
Yes
0.659
173
Yes
0.631
173
Yes
0.659
173
Yes
(11)
-0.0592
(0.3820)
-0.15
-0.0893
(0.0936)
-0.95
-0.0912
(0.0949)
-0.96
Ethnic Inequality in Population
[Gini Coeff., GREG]
adjusted R-squared
observations
Region Fixed Effects
(10)
-1.1868*** -1.1099** -1.2446** -1.3447*** -1.4580***
(0.4069) (0.5045) (0.4835) (0.5148)
(0.4905)
-2.92
-2.20
-2.57
-2.61
-2.97
-0.1112
(0.3780)
-0.29
Spatial Inequality 2
(8)
0.643
173
Yes
0.661
173
Yes
0.659
173
Yes
0.659
173
Yes
0.1060
(0.4155)
0.26
-0.1675
(0.1286)
-1.30
-0.1723
(0.1304)
-1.32
0.5164
(0.4472)
1.15
0.5305
(0.4465)
1.19
0.659
173
Yes
0.659
173
Yes
Table 2b - Baseline Estimates: Ethnic Inequality and Economic Development (in 2000), Ethnologue
(1)
Ethnic Inequality
[Gini Coeff., ETHNO]
(2)
-1.1281***
(0.2267)
-4.98
Spatial Inequality 1
(4)
-1.0245***
(0.2975)
-3.44
-1.1508***
(0.2786)
-4.13
[Gini Coeff., Pixels]
(3)
(5)
(6)
-1.0839***
(0.3817)
-2.84
(7)
-0.1857
(0.3617)
-0.51
-1.1612***
(0.2559)
-4.54
[Gini Coeff., Thiessen Polyg]
(9)
-0.0732
(0.4343)
-0.17
Log Number of Languages
[ETHNO]
-0.0884
(0.4354)
-0.20
-0.1730*** 0.0389
(0.0467) (0.0741)
-3.70
0.53
0.654
173
Yes
0.626
173
Yes
0.653
173
Yes
0.631
173
Yes
0.652
173
Yes
(11)
-0.1531
(0.3508)
-0.44
0.0357
(0.0759)
0.47
0.625
173
Yes
0.653
173
Yes
0.651
173
Yes
0.1967
(0.4539)
0.43
0.0399
(0.0751)
0.53
Ethnic Inequality in Population
[Gini Coeff., ETHNO]
adjusted R-squared
observations
Region Fixed Effects
(10)
-1.2806*** -1.1734** -1.2309*** -1.0657*** -1.2554***
(0.3694) (0.4625) (0.4528) (0.3747)
(0.4300)
-3.47
-2.54
-2.72
-2.84
-2.92
-0.2035
(0.3524)
-0.58
Spatial Inequality 2
(8)
0.651
173
Yes
0.6678*
(0.3727)
1.79
0.7013*
(0.3742)
1.87
0.66
173
Yes
0.661
173
Yes
The table reports cross-country OLS estimates. The dependent variable is the log of real GDP per capita in 2000. The ethnic Gini coefficients reflect inequality in lights per capita
across ethnic homelands. In Table 3A we use the digitized version of the Atlas Narodov Mira (GREG) to aggregate lights per capita across ethnic homelands. In Table 3B we use
the digitized version of the Ethnologue database to aggregate lights per capita across linguistic homelands.
The spatial Gini coefficient 1 captures the degree of spatial inequality across 2.5 by 2.5 decimal degree boxes/pixels in each country (boxes/pixels intersected by national
boundaries are of smaller size). The spatial Gini coefficient 2 captures the degree of spatial inequality across Thiessen polygons in each country. Thiessen polygons have the
unique property that each polygon contains only one input point, and any location within a polygon is closer to its associated point than to the point of any other polygon. The
input points are the centroids of the linguistic homelands according to the Ethnologue dataset. To construc the spatial inequality 2 index we intersect the 7,570 Thiessen polygons
with the country boundaries of 2000 and compute the spatial Gini across the resulting polygons within each country.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted
standard errors are reported in parentheses below the estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Table 3 - Sensitivity Checks A: Ethnic Inequality and Economic Development (in 2000)
Conditioning on Ethno-linguistic Fragmentation and Polarization
Atlas Narodov Mira (GREG)
(1)
Ethnic Inequality
[Gini Coeff.]
(2)
(3)
-0.0097
(0.3590)
-0.03
Cultural Fragmentation
(5)
(6)
(7)
(8)
0.1256
(0.3431)
0.37
-0.3417
(0.3520)
-0.97
Ethno-linguistic Polarization 1
0.0339
(0.3469)
0.10
0.464
(0.9745)
0.48
Ethno-linguistic Polarization 2
adjusted R-squared
observations
Region Fixed Effects
(4)
-1.4977*** -1.6771*** -1.5567*** -1.5782*** -1.1099*** -0.8248** -1.1186*** -1.1438***
(0.4011)
(0.3955)
(0.4075)
(0.3924)
(0.3716)
(0.3417)
(0.3744)
(0.3756)
-3.73
-4.24
-3.82
-4.02
-2.99
-2.41
-2.99
-3.04
Ethnic Fragmentation
Spatial Inequality 2
[Gini Coeff.]
Ethnologue
0.5817
(0.9820)
0.59
1.9997
(1.2688)
1.58
2.1173*
(1.1802)
1.79
0.0397
(0.4163)
0.10
0.0739
(0.4827)
0.15
0.0641
(0.3928)
0.16
0.0563
(0.3845)
0.15
-0.0862
(0.4408)
-0.20
-0.468
(0.4760)
-0.98
-0.0577
(0.4230)
-0.14
-0.0568
(0.4224)
-0.13
0.657
173
Yes
0.694
150
Yes
0.653
172
Yes
0.658
172
Yes
0.65
173
Yes
0.675
150
Yes
0.646
172
Yes
0.651
172
Yes
The table reports cross-country OLS estimates. The dependent variable is the log of real GDP per capita in 2000. In columns (1) and (5) we
control for ethnic fragmentation using an index that reflects the likelihood that two randomly chosen individuals in one country will be
members of the same group (from Alesina et al., 2003). In columns (2) and (6) we control for cultural (linguistic) fragmentation using an
index (from Fearon, 2003) that accounts for linguistic distances among ethnic groups. In columns (3) and (7) we control for ethnic
polarization, using the Montalvo and Reynal-Querol (2005) index. In columns (4) and (8) we control for ethnic polarization using a
polarization index that accounts for linguistic distances among ethnic groups (from Duclos, Esteban, and Rey (2004) and Esteban and Rey
(2011, 2012)).
In all specification we control for the overall degree of spatial inequality in a country using the Gini coefficient of lights per capita based on
Thiessen polygons that use as input points the centroids of the linguistic homelands according to the Ethnologue dataset. All specifications
include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources.
Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate statistical significance at
the 1%, 5%, and 10% level, respectively.
Table 4 - Sensitivity Checks B: Ethnic Inequality and Economic Development (in 2000)
Additional Controls and Alternative Measures of Ethnic Inequality
Atlas Narodov Mira (GREG)
All Ethnic Areas
(1)
(2)
Ethnic Inequality
[Gini Coeff.]
Excl. Capitals
(3)
(4)
Ethnologue
Excl. Small Groups
(5)
(6)
All Ethnic Areas
(7)
(8)
Excl. Capitals
(9)
(10)
Excl. Small Groups
(11)
(12)
-1.5123*** -1.3073*** -1.1601*** -1.0370*** -2.0069*** -1.5247*** -1.0650*** -0.8779*** -1.1723*** -0.8062** -1.3370***-1.2490***
(0.4357)
(0.3815) (0.4023) (0.3309) (0.5701) (0.5501)
(0.3480) (0.2997) (0.3747) (0.3552) (0.4800) (0.4535)
-3.47
-3.43
-2.88
-3.13
-3.52
-2.77
-3.06
-2.93
-3.13
-2.27
-2.79
-2.75
Spatial Inequality 2
[Gini Coeff.]
0.0743
(0.4710)
0.16
0.2679
(0.4043)
0.66
-0.7642
(0.5395)
-1.42
-0.275
(0.4501)
-0.61
-0.1669
(0.4365)
-0.38
-0.0063
(0.3968)
-0.02
-0.2035
(0.4908)
-0.41
0.0377
(0.4370)
0.09
-0.4793
(0.4822)
-0.99
-0.2985
(0.4757)
-0.63
-0.2473
(0.5105)
-0.48
0.1042
(0.4425)
0.24
Ethnic
Fragmentation
0.0943
(0.3716)
0.25
0.1826
(0.3162)
0.58
0.0709
(0.3904)
0.18
0.2411
(0.3437)
0.7
0.5099
(0.4040)
1.26
0.5054
(0.3667)
1.38
0.1308
(0.3614)
0.36
0.2437
(0.3134)
0.78
-0.1721
(0.3958)
-0.43
0.1237
(0.3544)
0.35
0.1909
(0.3578)
0.53
0.3108
(0.3151)
0.99
0.723
173
Yes
Rich
0.676
152
Yes
Simple
0.741
152
Yes
Rich
0.673
173
Yes
Simple
0.723
173
Yes
Rich
0.654
173
Yes
Simple
0.715
173
Yes
Rich
0.672
147
Yes
Simple
0.723
147
Yes
Rich
0.654
173
Yes
Simple
0.7173
173
Yes
Rich
adjusted R-squared
0.663
Observations
173
Region Fixed Effects Yes
Controls
Simple
The table reports cross-country OLS estimates. The dependent variable is the log of real GDP per capita in 2000. In columns (1)-(6) we use the digitized version of the Atlas
Narodov Mira (GREG) to aggregate lights per capita across ethnic homelands and construct the ethnic inequality measures. In columns (7)-(12) we use the digitized version of the
Ethnologue database to aggregate lights per capita across linguistic homelands and construct the ethnic inequality measures. For the construction of the ethnic inequality measures
(Gini coefficients) in columns (3), (4), (9), and (10) we exclude ethnic areas where capital cities fall. For the construction of the ethnic inequality measures (Gini coefficients) in
columns (5), (6), (11), and (12) we exclude small ethnic groups consisting of less than one percept of country’s population.
Odd-numbered columns include as controls absolute latitude, log land area, and log population in 2000 (simple set of controls). Even-numbered columns include as controls
absolute latitude, log land area, log population in 2000, an index of terrain ruggedness, the percentage of each country with fertile soil, the percentage of each country with tropical
climate, average distance to nearest ice-free coast, and an index of gem-quality diamond extraction (rich set of controls). In all specifications we control for ethnic fragmentation
using an index that reflects the likelihood that two randomly chosen individuals in one country will be members of the same group (from Alesina et al., 2003).
In all specification we control for the overall degree of spatial inequality in a country using the Gini coefficient of lights per capita based on Thiessen polygons that use as input
points the centroids of the linguistic homelands according to the Ethnologue dataset. All specifications include continental fixed effects (constants not reported). The Data
Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Table 5. The Origins of Contemporary Ethnic Inequality
Inequality in Geographic Endowments across Ethnic Homelands and Contemporary Ethnic Inequality in Development across Ethnic
Atlas Narodov Mira (GREG)
All Ethnic Areas
(1)
(2)
(3)
(4)
Ethnologue
Excl. Capitals
(5)
(6)
All Ethnic Areas
(7)
(8)
(9)
Excl. Capitals
(10)
(11)
(12)
Gini Coef. - Sea Distance
0.0489
(0.0956)
0.1379
(0.1231)
0.1857
(0.1156)
0.0552
(0.1274)
0.1090
(0.1611)
0.1961
(0.1494)
0.0355
(0.0969)
-0.1886
(0.1567)
-0.1756
(0.1622)
0.0897
(0.1143)
Gini Coef. - Elevation
0.4343
(0.3304)
0.8727
(0.6673)
0.6973
(0.6334)
0.0953
(0.1206)
0.1046
(0.1262)
0.0754
(0.1178)
0.6292
(0.4782)
3.0353**
(1.2705)
1.7750
-1.2926
0.0027** 0.0038*** 0.0019
(0.0010) (0.0011) (0.0012)
Gini Coeff. -Land Quality 0.2528*** -0.1028
(0.0910) (0.1775)
-0.121 0.3067*** 0.1227
(0.1813) (0.0935) (0.1816)
0.0276
(0.1639)
0.0231
(0.1481)
0.1227 0.3680*** 0.6496*** 0.6000*** 0.3338*** 0.4759** 0.4116**
(0.1872) (0.0978) (0.2038) (0.1654) (0.0922) (0.1873) (0.1658)
Gini Coeff. - Water Area 0.5632*** 0.4985*** 0.4744*** 0.4476*** 0.4011*** 0.3914*** 0.6698*** 0.5550*** 0.5862*** 0.5238*** 0.4206*** 0.4330***
(0.0550) (0.0773) (0.0733) (0.0970) (0.1275) (0.1280) (0.0621) (0.0857) (0.0875) (0.0786) (0.0997) (0.0934)
adjusted R-squared
Observations
Region Fixed Effects
Controls
0.674
173
0.6674
169
0.6408
162
0.5967
151
0.5988
150
0.6078
150
0.7511
168
0.7537
166
0.7595
160
0.7169
144
0.714
143
0.7398
142
Yes
No
Yes
Spatial
Yes
Spatial &
Level
Yes
No
Yes
Spatial
Yes
Spatial &
Level
Yes
No
Yes
Spatial
Yes
Spatial &
Level
Yes
No
Yes
Spatial
Yes
Spatial &
Level
The table reports cross-country OLS estimates, associating contemporary ethnic inequality with inequality in geographic endowments across ethnic homelands. The dependent
variable is the ethnic Gini coefficient that reflects inequality in lights per capita across ethnic-linguistic homelands, using the digitized version of Atlas Narodov Mira (GREG) in (1)(6) and Ethnologue in (7)-(12). To construct the inequality measures in geographic endowments we first estimate the distance of the centroid of each ethnic homeland to the closest
sea coast, average elevation, average soil quality, and the area of each homeland covered by water (lakes, rivers, and other streams) and then construct Gini coefficients capturing
inequality in each of these geographic features for each country.
In columns (1)-(3) and (7)-(9) we use all ethnic-linguistic homelands; in columns (4)-(6) and (10)-(12) we exclude ethnic-linguistic regions where capital cities fall. In columns (2),
(5), (8), and (11) we control for the overall degree of spatial inequality in geographic endowments using the Gini coefficient of each of these features (distance to the closest sea
coast, elevation, soil quality, water area) based on Thiessen polygons that use as input points the centroids of the linguistic homelands according to the Ethnologue dataset. In
columns (3), (6), (9), and (12) we also control for the mean value of distance to closest sea coast, elevation, soil quality, and area under water. All specifications include continental
fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses
below the estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Table 6 - Principal Component Analysis
Eigenvalue
Variance
Explained
Factor Loadings
Variable
1st PC
2nd PC
3rd PC
4th PC
0.080
-0.720
0.689
-0.027
0.563
0.208
0.120
-0.791
0.153
0.468
0.230
-0.840
0.720
-0.650
0.152
-0.190
0.186
-0.770
0.610
0.008
0.563
0.229
0.128
-0.783
-0.468
0.456
0.722
-0.228
0.627
-0.095
0.234
-0.737
Panel A: Gini Coefficient GREG - All Groups (173 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
2.474
0.708
0.481
0.337
0.618
0.177
0.120
0.084
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.482
0.491
0.479
0.545
-0.666
0.444
0.531
-0.277
Panel B: Gini Coefficient GREG - Excluding Capitals (151 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
2.420
0.753
0.515
0.312
0.567
0.200
0.128
0.105
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.537
0.540
0.402
0.508
-0.413
-0.259
0.873
0.020
Panel C: Gini Coefficient ETHNOLOGUE - All Groups (168 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
2.389
0.739
0.583
0.289
0.597
0.185
0.146
0.072
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.498
0.478
0.445
0.571
-0.633
0.355
0.643
-0.246
Panel D: Gini Coefficient ETHNOLOGUE - Excluding Capitals (144 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
1.804
1.052
0.703
0.441
0.451
0.263
0.176
0.110
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.594
0.186
0.466
0.629
0.188
0.865
-0.455
-0.096
Panel F: Gini Coefficient - Overall Spatial Inequality Index 1 - Pixel 2.5 x 2.5 degrees (164 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
2.132
0.814
0.580
0.474
0.533
0.203
0.145
0.118
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.502
0.497
0.462
0.536
-0.559
0.404
0.616
-0.381
0.131
-0.759
0.637
0.032
0.647
0.117
0.045
-0.752
Panel E: Gini Coefficient - Overall Spatial Inequality Index 2 - Thiessen Polygons (169 countries)
1st Principal Component
2nd Principal Component
3rd Principal Component
4th Principal Component
2.082
0.849
0.598
0.471
0.521
0.212
0.150
0.118
Gini Sea Distance
Gini Elevation
Gini Land Quality
Gini Water Area
0.480
0.515
0.451
0.549
-0.608
0.347
0.639
-0.319
0.377
-0.662
0.612
-0.212
0.507
0.420
-0.116
-0.744
The table reports the results of the principal component analysis that is based on four measures (Gini coefficients) reflecting inequality in
geographic endowments in distance to the coast, elevation, land suitability for agriculture, and area under water across ethnic homelands
(Panels A, B, C, and D), pixels of 2.5 x 2.5 decimal degrees (in Panel E) and Thiessen polygons that use as input points the centroids of the
linguistic homelands according to the Ethnologue dataset (Panel F). Column (1) reports the eigenvalue of each principal component and
column (2) gives the percentage of the total variance explained by each principal component. The other columns give the factor loadings in
the four principal components of the Gini coefficient reflecting inequality in distance to the coast, elevation, land suitability for agriculture,
and area under water.
Table 7: The Origins of Contemporary Ethnic Inequality
Inequality in Geographic Endowments across Ethnic Homelands and Contemporary Ethnic Inequality
Atlas Narodov Mira (GREG)
(1)
(2)
(3)
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
0.1227*** 0.1089*** 0.0973***
(0.0088)
(0.0178)
(0.0171)
14.00
Spatial Inequality in Geographic
Endowments (PC)
Adjusted R-squared
Observations
Region Fixed Effects
Additional Controls
0.623
173
Yes
No
6.13
5.70
0.0109
(0.0182)
0.6
0.0068
(0.0194)
0.35
0.616
169
Yes
No
0.605
162
Yes
Geography
(4)
Ethnologue
(5)
(6)
0.1613*** 0.1523*** 0.1588***
(0.0095) (0.0196)
(0.0199)
16.99
0.695
168
Yes
No
7.76
7.98
0.0075
(0.0202)
0.37
-0.006
(0.0228)
-0.26
0.689
166
Yes
No
0.688
160
Yes
Geography
The table reports cross-country OLS estimates, associating contemporary ethnic inequality with inequality in geographic endowments
across ethnic homelands. The dependent variable is the ethnic Gini coefficient that reflects inequality in lights per capita across
ethnic-linguistic homelands in 2000, using the digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue
(in columns (4)-(6)).
The main independent variable is a composite index capturing inequality in geographic endowments across ethnic homelands. The
index is the first principal component of inequality across ethnic-linguistic homelands in distance to the coast, elevation, land
suitability for agriculture, and area under water. In columns (2), (3), (5), and (6) we control for the overall degree of spatial inequality
in geographic endowments using a composite index that aggregates (via principal components) Gini coefficients on distance to the
coast, elevation, land suitability for agriculture, water area across Thiessen polygons that use as input points the centroids of the
linguistic homelands according to the Ethnologue dataset. In columns (3) and (6) we also control for the mean value of distance to
the coast, elevation, land suitability for agriculture, and area under water.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions
and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Table 8: Inequality in Geographic Endowments across Ethnic Homelands and Contemporary
Development
Atlas Narodov Mira (GREG)
(1)
(2)
(3)
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-0.1789*** -0.2311*** -0.1770**
(0.0405)
(0.0851)
(0.0891)
-4.42
-2.71
-1.99
Spatial Inequality in Geographic
Endowments (PC)
adjusted R-squared
Observations
Region Fixed-Effects
Additional Controls
0.629
173
Yes
No
0.0755
(0.0970)
0.78
0.048
(0.1113)
0.43
0.646
169
Yes
No
0.668
162
Yes
Geography
(4)
Ethnologue
(5)
(6)
-0.1611*** -0.058
(0.0459) (0.0956)
-3.51
-0.61
-0.1526*
(0.0860)
-1.78
-0.0898
(0.0986)
-0.91
0.0268
(0.1093)
0.24
0.639
166
Yes
No
0.673
160
Yes
Geography
0.623
168
Yes
No
The table reports cross-country OLS estimates, associating contemporary economic development with inequality in geographic
endowments across ethnic homelands. The dependent variable is the log of real GDP per capita in 2000.
The main independent variable is a composite index capturing inequality in geographic endowments across ethnic homelands, using
the digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue (in columns (4)-(6)). The index is the first
principal component of inequality across ethnic-linguistic homelands in distance to the coast, elevation, land suitability for
agriculture, and area under water. In columns (2), (3), (5), and (6) we control for the overall degree of spatial inequality in
geographic endowments using a composite index that aggregates (via principal components) Gini coefficients on distance to the
coast, elevation, land suitability for agriculture, water area across Thiessen polygons that use as input points the centroids of the
linguistic homelands according to the Ethnologue dataset. In columns (3) and (6) we also control for the mean value of distance to
the coast, elevation, land suitability for agriculture, and area under water.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions
and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Table 9 - Ethnic Inequality and Regional Development within African Countries
District-level and Individual-Level Analysis Using Data from the Afrobarometer Surveys
District-Level
(1)
(2)
(3)
Individual-Level
(4)
(5)
(6)
Panel A: Living Conditions
Between Group Ineq.
[Theil Index]
Within Group Ineq.
[Theil Index]
adjusted R-squared
Observations
-5.4994*** -6.2017***
(0.8809)
(1.0308)
-5.9472*** -5.8051*** -4.7228*** -5.1732***
(1.0874)
(0.8644)
(0.8488)
(0.8604)
-2.6220**
(0.9085)
-5.7490***
(1.0771)
-2.3695*** -5.4641*** -1.4248*** -4.5081***
(0.4606)
(0.5605)
(0.3351)
(0.5255)
0.356
1298
0.500
842
0.154
20617
0.166
16580
0.188
20617
0.188
16580
Panel B: Education
Between Group Ineq.
[Theil Index]
-4.4948*** -4.7501***
(0.9452)
(1.2602)
-3.2780*** -2.7169** -3.4919*** -3.0329***
(1.0931)
(1.1152)
(1.0357)
(1.0617)
Within Group Ineq.
[Theil Index]
-3.5337*** -3.9965***
(0.5205)
(0.8417)
-1.5350*** -1.8955*** -1.3886*** -2.0192***
(0.5110)
(0.6833)
(0.5003)
(0.6909)
adjusted R-squared
Observations
Country Fixed Effects
Country-Ethnicity Fixed Effects
District-level Controls
Individual-level Controls
Sample; Districts
0.506
1298
0.547
842
0.484
20617
0.483
16580
0.498
20617
0.495
16580
Yes
No
Yes
No
All
Yes
No
Yes
No
>1 group
Yes
No
Yes
Yes
All
Yes
No
Yes
Yes
>1 group
No
Yes
Yes
Yes
All
No
Yes
Yes
Yes
>1 group
The table reports OLS estimates, associating two proxies of well-being (living conditions in Panel A and education in Panel B) with
inequality between and within ethnic groups at the district level, as reflected in the Theil index. The dependent variable in Panel A is a 15 living conditions index; the dependent variable in Panel B is a 1-10 education index. Columns (1)-(2) report district-level estimates,
while columns (3)-(8) report individual level estimates. The between-ethnic-group and the within-ethnic-group Theil indicators are based
on individuals’ responses on living conditions. The district-level conditioning set includes the log number of ethnic groups in each
district, the log number of respondents in each district, and district-level ethnic fractionalization. The individual-level conditioning set
includes age, age squared, a gender indicator variable, 22 religion fixed effects and 25 occupation fixed effects. Odd-numbered columns
report estimates in the full sample. In even-numbered columns we exclude from the estimation districts with respondents from just one
ethnic group. All specifications include country fixed effects (constants not reported). Specifications (5) and (6) include ethnicity-country
fixed effects (constants not reported).
The Data Appendix gives detailed variable definitions and data sources. All variables are constructed using data from the 3rd round of the
Afrobarometer Surveys. Clustered at the district level standard errors are reported in parentheses below the estimates. ***, **, and *
indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Table 10a - Ethnic Inequality and Public Goods Provision within African Countries
Piped Water
District
(1)
Between Group Ineq.
[Theil Index]
Within Group Ineq.
[Theil Index]
adjusted R-squared
Observations
Country Fixed Effects
Country-Ethnicity Fixed Effects
District Controls
Individual Controls
Sample, Districts
Sewage System
Individual Level
(2)
(3)
-2.5917** -1.9828***-1.8395***
(0.9664) (0.6336) (0.5793)
-0.9725**
(0.3637)
-0.4500
(0.3774)
District
(4)
Electrification
Individual Level
(5)
(6)
-1.9295*** -1.3094** -1.3981***
(0.4999)
(0.5217)
(0.5135)
District
(7)
Individual Level
(8)
(9)
-1.7328**
(0.7864)
-1.1828*
(0.6140)
-1.2205**
(0.5995)
-0.4652
(0.3563)
-0.4209
(0.5185)
0.058
(0.3194)
0.1477
(0.3325)
-1.2173***
(0.2345)
-0.2566
(0.3421)
-0.4046
(0.3107)
0.238
0.392
0.441
837
16378
16378
Yes
Yes
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
>1 groups >1 groups >1 groups
0.219
833
Yes
No
Yes
Yes
>1 groups
0.399
16067
Yes
No
Yes
Yes
>1 groups
0.421
16067
No
Yes
Yes
Yes
>1 groups
0.307
839
Yes
No
Yes
Yes
>1 groups
0.44
16429
Yes
No
Yes
Yes
>1 groups
0.478
16429
No
Yes
Yes
Yes
>1 groups
Table 10b - Ethnic Inequality and Public Goods Provision within African Countries
Piped Water
District
(1)
Between Group Ineq.
[Theil Index]
Within Group Ineq.
[Theil Index]
Mean Living Conditions at the District
adjusted R-squared
Observations
Country Fixed Effects
Country-Ethnicity Fixed Effects
District Controls
Individual Controls
Sample, Districts
Sewage System
Individual Level
(2)
(3)
-2.1828* -1.8972***-1.8137***
(1.0963) (0.6526) (0.5900)
District
(4)
Electrification
Individual Level
(5)
(6)
-1.7909*** -1.3979*** -1.4289***
(0.5177)
(0.5370)
(0.5290)
District
(7)
Individual Level
(8)
(9)
-1.3118
(0.8746)
-1.3686**
(0.6350)
-1.2675**
(0.6173)
-0.5935*
(0.3305)
-0.3748
(0.4178)
-0.4443
(0.3875)
-0.2897
(0.5116)
-0.0195
(0.3330)
0.1229
(0.3385)
-0.8267***
(0.1815)
-0.4204
(0.3633)
-0.4431
(0.3399)
0.0659
(0.0451)
0.0137
(0.0307)
0.0043
(0.0288)
0.0228
(0.0243)
-0.0143
(0.0229)
-0.0052
(0.0240)
0.0678*
(0.0332)
-0.0297
(0.0302)
-0.0079
(0.0269)
0.238
0.392
0.441
837
16378
16378
Yes
Yes
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
Yes
>1 groups >1 groups >1 groups
0.219
833
Yes
No
Yes
Yes
>1 groups
0.399
16067
Yes
No
Yes
Yes
>1 groups
0.421
16067
No
Yes
Yes
Yes
>1 groups
0.307
839
Yes
No
Yes
Yes
>1 groups
0.44
16429
Yes
No
Yes
Yes
>1 groups
0.478
16429
No
Yes
Yes
Yes
>1 groups
The table reports OLS estimates, associating measures of public goods (access to piped water, access to a sewage system, and access to an electricity grid) with inequality between
and within ethnic groups at the district level, as reflected in the Theil index. Columns (1), (4), and (7) report district-level estimates, while columns (2), (3), (5), (6), (8), and (9) report
individual level estimates. The dependent variable in columns (1)-(3) is a dummy variable that takes on the value of one for households that have access to clean piped water; in
columns (4)-(6) is a dummy variable that takes on the value of one for households that have access to a sewage system; in columns (7)-(9) is a dummy variable that takes on the value
of one for households that have access to an electricity grid. The between-ethnic-group and the within-ethnic-group Theil indicators are based on individuals’ responses on living
conditions.
The district-level conditioning set includes the log number of ethnic groups in each district, the log number of respondents in each district, and district-level ethnic fractionalization.
The individual-level conditioning set includes age, age squared, a gender indicator variable, 22 religion fixed effects, 25 occupation fixed effects, 5 living conditions fixed effects, and
9 education fixed effects. In all specifications we exclude from the estimation districts with respondents from just one ethnic group. Panel B reports otherwise identical to Panel A
specifications including also in the set of explanatory variables the mean value of the living conditions index in the district. All specifications include country fixed effects (constants
not reported). Specifications (3), (6), and (9) include ethnicity-country fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources.
All variables are constructed using data from the 3rd round of the Afrobarometer Surveys. Clustered at the district level standard errors are reported in parentheses below the
estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Table 11a - Ethnic Inequality and Beliefs on Democracy within African Countries
Support for Democracy
District
(1)
Between Group Ineq.
[Theil Index]
Within Group Ineq.
[Theil Index]
adjusted R-squared
Observations
Country Fixed Effects
Country-Ethnicity Fixed Effects
District Controls
Individual Controls
Sample, Districts
Individual Level
(2)
(3)
Satisfaction with Democracy
District
(4)
Individual Level
(5)
(6)
Constituion Reflects People's Values-Hopes
District
(7)
Individual Level
(8)
(9)
-1.4171*
(0.7702)
-0.6833
(0.6387)
-0.9821
(0.6341)
-2.5130** -1.9787*** -1.8511***
(0.8794)
(0.7619)
(0.6824)
-1.8538*
(0.9479)
-1.6343*
(0.8803)
-1.4638*
(0.8169)
-0.5123
(0.3996)
-0.5371*
(0.3228)
-0.4089
(0.3101)
-1.3277**
(0.4655)
-1.2481**
(0.5182)
-0.9595**
(0.4654)
-0.0716
(0.8013)
-0.6032
(0.5818)
0.1742
(0.4933)
0.244
838
Yes
No
Yes
No
>1 group
0.085
14381
Yes
No
Yes
Yes
>1 group
0.098
14381
No
Yes
Yes
Yes
>1 group
0.414
837
Yes
No
Yes
No
>1 group
0.156
14231
Yes
No
Yes
Yes
>1 group
0.184
14231
No
Yes
Yes
Yes
>1 group
0.187
839
Yes
No
Yes
No
>1 group
0.052
13923
Yes
No
Yes
Yes
>1 group
0.071
13923
No
Yes
Yes
Yes
>1 group
Table 11b - Ethnic Inequality and Beliefs on Democracy within African Countries
Support for Democracy
District
(1)
Between Group Ineq.
Individual Level
(2)
(3)
Satisfaction with Democracy
District
(4)
Individual Level
(5)
(6)
Constitution Reflects People's Values-Hope
District
(7)
Individual Level
(8)
(9)
-1.2924*
(0.6844)
-0.6556
(0.6489)
-1.0286
(0.6442)
-1.7639*
(0.9030)
-1.181
(0.7817)
-1.5854**
(0.6962)
-0.933
(0.9094)
-0.4923
(0.9260)
-0.8835
(0.8553)
-0.3964
(0.3520)
-0.5134
(0.3497)
-0.4458
(0.3268)
-0.6305
(0.5149)
-0.5556
(0.5320)
-0.7441
(0.4956)
0.7744
(0.7115)
0.3774
(0.5962)
0.6381
(0.5442)
Mean Living Conditions at the District
0.0202
(0.0301)
0.0044
(0.0257)
-0.0077
(0.0260)
0.1215**
(0.0429)
0.1271***
(0.0386)
0.0445
(0.0343)
0.1492***
(0.0427)
0.1756***
(0.0482)
0.0932**
(0.0410)
adjusted R-squared
Observations
Country Fixed Effects
Country-Ethnicity Fixed Effects
District Controls
Individual Controls
Sample, Districts
0.243
838
Yes
No
Yes
Yes
>1 group
0.085
14381
Yes
No
Yes
Yes
>1 group
0.098
14381
No
Yes
Yes
Yes
>1 group
0.421
837
Yes
No
Yes
Yes
>1 group
0.157
14231
Yes
No
Yes
Yes
>1 group
0.184
14231
No
Yes
Yes
Yes
>1 group
0.2
839
Yes
No
Yes
Yes
>1 group
0.054
13923
Yes
No
Yes
Yes
>1 group
0.072
13923
No
Yes
Yes
Yes
>1 group
[Theil Index]
Within Group Ineq.
[Theil Index]
The table reports OLS estimates, associating measures reflecting individual’s beliefs on support and satisfaction with democracy with inequality between and within ethnic groups at the
district level, as reflected in the Theil index. Columns (1), (4), and (7) report district-level estimates, while columns (2), (3), (5), (6), (8), and (9) report individual level estimates. The
dependent variable in columns (1)-(3) is a trichotomous (0-3 range) variable that reflects’ individual’s beliefs on support for democratic rule; in columns (4)-(6) the dependent variable is
an ordered (0-4 range) index that reflects individual’s satisfaction with the functioning of democratic institutions; in columns (7)-(9) the dependent variable is an ordered (1-5 range) index
that captures individual’s beliefs on whether the constitution reflects people’s hope and values. The between-ethnic-group and the within-ethnic-group Theil indicators are based on
individuals’ responses on living conditions. The district-level conditioning set includes the log number of ethnic groups in each district, the log number of respondents in each district, and
district-level ethnic fractionalization. The individual-level conditioning set includes age, age squared, a gender indicator variable, 22 religion fixed effects, 25 occupation fixed effects, 5
living conditions fixed effects, and 9 education fixed effects.
In all specifications we exclude from the estimation districts with respondents from just one ethnic group. Panel B reports otherwise identical to Panel A specifications including also in the
set of explanatory variables the mean value of the living conditions index in the district.All specifications include country fixed effects (constants not reported). Specifications (3), (6), and
(9) include ethnicity-country fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources. All variables are constructed using data from
the 3rd round of the Afrobarometer Surveys. Clustered at the district level standard errors are reported in parentheses below the estimates. ***, **, and * indicate statistical significance at
the 1%, 5%, and 10% level, respectively.
Appendix Table 1: Correlation Structure - Cross-Country Measures
Panel A: Ethnic Inequality Indicators (all ethnic areas)
Ethnic Inequality Indicators - Gini Coefficients
GREG
Ethnologue
Overall Spatial Inequality Indicators - Gini Coefficients
Spatial Gini 1
Spatial Gini 2
Ethnic Gini 2009 (GREG)
Ethnic Gini 2000 (GREG)
Ethnic Gini 1992 (GREG)
1
0.9545*
0.9382*
1
0.9519*
1
Ethnic Gini 2009 (ETHN)
Ethnic Gini 2000 (ETHN)
Ethnic Gini 1992 (ETHN)
0.7679*
0.7625*
0.7719*
0.7564*
0.7619*
0.7686*
0.7686*
0.7696*
0.7958*
1
0.9914*
0.9759*
1
0.9775*
1
Spatial Gini 2009 (Thiessen) 0.7784*
Spatial Gini 2000 (Thiessen) 0.7724*
Spatial Gini 1992 (Thiessen) 0.7870*
0.7760*
0.7903*
0.7986*
0.7755*
0.7777*
0.8136*
0.8180*
0.8053*
0.8242*
0.8205*
0.8218*
0.8289*
0.8082*
0.8012*
0.8397*
1
0.9835*
0.9639*
1
0.9604*
1
Spatial Gini 2009 (Pixels)
Spatial Gini 2000 (Pixels)
Spatial Gini 1992 (Pixels)
0.7002*
0.7190*
0.7187*
0.6633*
0.6642*
0.6886*
0.7161*
0.7129*
0.7347*
0.7214*
0.7273*
0.7446*
0.7096*
0.7066*
0.7535*
0.7699*
0.7710*
0.7807*
0.7845*
0.8002*
0.7970*
0.7638*
0.7676*
0.8129*
0.6826*
0.6851*
0.6860*
1
0.9736*
0.9384*
1
0.9463*
1
Appendix Table 1: Correlation Structure - Cross-Country Measures
Panel B: Ethnic Inequality Indicators (excl. capitals)
Ethnic Inequality Indicators - Gini Coefficients
GREG
Ethnologue
Overall Spatial Inequality Indicators - Gini Coefficients
Spatial Gini 1
Spatial Gini 2
Ethnic Gini 2009 (GREG)
Ethnic Gini 2000 (GREG)
Ethnic Gini 1992 (GREG)
1
0.9440*
0.8965*
1
0.9209*
1
Ethnic Gini 2009 (ETHN)
Ethnic Gini 2000 (ETHN)
Ethnic Gini 1992 (ETHN)
0.6673*
0.6735*
0.6399*
0.6748*
0.6942*
0.6627*
0.6438*
0.6507*
0.6918*
1
0.9905*
0.9533*
1
0.9521*
1
Spatial Gini 2009 (Thiessen) 0.6750*
Spatial Gini 2000 (Thiessen) 0.6788*
Spatial Gini 1992 (Thiessen) 0.6842*
0.6898*
0.7183*
0.7189*
0.6205*
0.6356*
0.6672*
0.8343*
0.8160*
0.8385*
0.8405*
0.8368*
0.8503*
0.7942*
0.7843*
0.8360*
1
0.9835*
0.9639*
1
0.9604*
1
Spatial Gini 2009 (Pixels)
Spatial Gini 2000 (Pixels)
Spatial Gini 1992 (Pixels)
0.6319*
0.6410*
0.6494*
0.5494*
0.5475*
0.5657*
0.6884*
0.6746*
0.7152*
0.7013*
0.6960*
0.7333*
0.6513*
0.6434*
0.7038*
0.7699*
0.7710*
0.7807*
0.7845*
0.8002*
0.7970*
0.7638*
0.7676*
0.8129*
0.6118*
0.6023*
0.6117*
1
0.9736*
0.9384*
1
0.9463*
1
Appendix Table 1: Correlation Structure - Cross-Country Measures
Panel C: Correlation of Ethnic Inequality Indicators with Measures of Ethnic-Linguistic-Religious Fragmentation, Polarization, and
Segregation
Ethnic Gini 2000 - All (GREG) 1
Ethnic Gini 2000 - All (ETHN)0.7619*
Spatial Gini 2000 (Thiessen) 0.7903*
Spatial Gini 2000 (Pixels)
0.7190*
Ethnic Fragmentation
0.4464*
Linguistic Fragmentation
0.3878*
Religious Fragmentation
-0.057
Ethnic Segregation
0.2944*
Linguistic Segregation
0.2437*
Religious Segregation
0.2552*
Ethnic Polarization 1
0.1042
Ethnic Polarization 2
0.0497
1
0.8218*
0.7273*
0.4666*
0.4123*
-0.0035
0.4468*
0.3711*
0.2449*
0.0955
0.0335
1
0.8002*
0.5099*
0.4653*
0.044
0.3348*
0.2266*
0.2249*
0.0144
-0.0345
1
0.4640*
0.3506*
0.0041
0.2064*
0.2131*
0.2097
0.0942
0.0254
1
0.6885*
0.1629*
0.4813*
0.3945*
0.2502*
0.3065*
0.1697*
1
0.2748*
0.3705*
0.3056*
0.2911*
0.2617*
0.1032
1
-0.0442
-0.0363
0.0811
-0.1019
-0.0389
1
0.8422*
0.2205
0.1196
0.0654
1
0.1276
0.1781
0.1151
1
0.0251
-0.0012
1
0.5161*
1
Appendix Table 1: Correlation Structure - Cross-Country Measures
Panel D: Correlation of Ethnic Inequality Indicators with Measures of Development and Income Inequality
Ethnic Gini - All (GREG)
1.0000
Ethnic Gini - Excl. Capitals (GREG)
0.9404*
Ethnic Gini - Excl. Small (GREG)
0.6992*
1.0000
0.6137*
1.0000
Ethnic Gini - All (ETHN)
0.7619*
Ethnic Gini - Excl. Capitals (ETHN)
0.7229*
Ethnic Gini - Excl. Small (ETHN)
0.6337*
0.7096*
0.6942*
0.6069*
0.6666*
0.6560*
0.7785*
1
0.9831*
0.8183*
1
0.7687*
1
Spatial Gini 2000 (Thiessen) 0.7903*
Spatial Gini 2000 (Pixels)
0.7190*
Income Inequality (Gini coeff.)0.2643*
0.7183*
0.6410*
0.2608*
0.7104*
0.5593*
0.3063*
0.8218*
0.7273*
0.3260*
0.8368*
0.6960*
0.3151*
0.7914*
0.6448*
0.4228*
Log real GDP p.c. in 2000
-0.5294*
Rule of Law in 2000
-0.4933*
Control of Corruption in 2000-0.4570*
-0.5193* -0.6552*
-0.5324* -0.5081*
-0.4944* -0.4984*
-0.4950* -0.5250* -0.5795*
-0.4464* -0.4869* -0.4999*
-0.4207* -0.4447* -0.4753*
1
0.8002*
0.3452*
1
0.2887*
1
-0.5611* -0.4556* -0.3751*
-0.4982* -0.4108* -0.3998*
-0.4658* -0.3677* -0.4041*
1
0.7952*
0.7400*
1
0.9423*
The table reports the correlation structure between the main variables employed in the cross-country empirical analysis. Panel A gives the correlation between the main ethnic
inequality measures and the overall spatial inequality measures in 1992, 2000, and 2009.
Panel B gives the correlation between ethnic inequality and the overall spatial inequality measures in 1992, 2000, and 2009. For the estimation of the ethnic inequality measures
(Gini coefficients) we exclude ethnic regions where capital cities fall.
Panel C gives the correlation between the main ethnic inequality measures and the overall spatial inequality measures in 2000 with measures reflecting ethnic, linguistic, and
religious fragmentation, segregation, and polarization.
Panel D gives the correlation between the main ethnic inequality measures and the overall spatial inequality measures in 2000 with income inequality and measures capturing
economic and institutional development in 2000.
The Data Appendix gives detailed variable definitions and data sources. * indicates statistical significance at the 5% level.
1
Appendix Table 2 - Additional Sensitivity Checks: Ethnic Inequality and Economic Development (in 2000)
Excluding Countries with One Ethnic-Linguistic Group
Panel A: Conditioning on Overall Spatial Inequality Index based on Thiessen Polygons
Atlas Narodov Mira (GREG)
All Ethnic Areas
Excl. Capitals
Excl. Small Groups
(1)
(2)
(3)
(4)
(5)
(6)
Ethnic Inequality
[Gini Coeff.]
Spatial Inequality 2
All Ethnic Areas
(7)
(8)
Ethnologue
Excl. Capitals
(9)
(10)
Excl. Small Groups
(11)
(12)
-1.6096*** -1.3812*** -1.1601*** -1.0370*** -2.0382*** -1.4606*** -1.1754*** -0.7805** -1.1723*** -0.8062** -1.3301*** -0.9689*
(0.4575) (0.3835) (0.4023) (0.3309) (0.5897) (0.5397)
(0.3749) (0.3514) (0.3747) (0.3552) (0.5041) (0.4964)
[Gini Coeff.]
-0.4639
(0.5269)
-0.0641
(0.4456)
-0.7642
(0.5395)
-0.275
(0.4501)
-0.8093*
(0.4791)
-0.4447
(0.4388)
-0.4912
(0.4748)
-0.3364
(0.4669)
-0.4793
(0.4822)
-0.2985
(0.4757)
-0.6475
(0.5049)
-0.3788
(0.4473)
Ethnic
Fragmentation
0.0863
(0.3834)
0.2430
(0.3359)
0.0709
(0.3904)
0.2411
(0.3437)
0.4818
(0.4254)
0.534
(0.3974)
-0.1566
(0.3721)
0.1251
(0.3373)
-0.1721
(0.3958)
0.1237
(0.3544)
-0.1395
(0.3780)
0.1456
(0.3440)
adjusted R-squared
Observations
Region Fixed Effects
Controls
0.680
153
Yes
Simple
0.741
153
Yes
Rich
0.676
152
Yes
Simple
0.741
152
Yes
Rich
0.691
153
Yes
Simple
0.739
153
Yes
Rich
0.677
148
Yes
Simple
0.729
148
Yes
Rich
0.672
147
Yes
Simple
0.723
147
Yes
Rich
0.674
148
Yes
Simple
0.7288
148
Yes
Rich
Appendix Table 2 - Additional Sensitivity Checks: Ethnic Inequality and Economic Development (in 2000), cont.
Excluding Countries with One Ethnic-Linguistic Group
Panel B: Conditioning on Overall Spatial Inequality Index based on Pixels of same Size (2.5 x 2.5 degrees)
Atlas Narodov Mira (GREG)
All Ethnic Areas
Excl. Capitals
Excl. Small Groups
(1)
(2)
(3)
(4)
(5)
(6)
Ethnic Inequality
[Gini Coeff.]
All Ethnic Areas
(7)
(8)
Ethnologue
Excl. Capitals
(9)
(10)
Excl. Small Groups
(11)
(12)
-1.3891*** -1.0916*** -1.1161*** -0.9068*** -1.9173*** -1.2843** -1.0929*** -0.7285** -1.1124*** -0.7595*** -1.2700*** -0.8987**
(0.4090) (0.3465) (0.3490) (0.2934) (0.5602) (0.5102)
(0.3121) (0.2925) (0.3027) (0.2829) (0.3931) (0.4043)
Spatial Inequality 1 -1.1909** -0.9672** -1.5238*** -1.2222*** -1.3685*** -1.1437*** -1.3388*** -1.0038** -1.3646*** -1.0080** -1.3792*** -0.9862**
[Gini Coeff.]
(0.5053)
(0.4173)
(0.4953)
(0.4028)
(0.4729)
(0.4032)
(0.4747)
(0.4440)
(0.4659)
(0.4347)
(0.4688)
(0.4195)
Ethnic
Fragmentation
0.1211
(0.3782)
0.2454
(0.3285)
0.1247
(0.3804)
0.2408
(0.3273)
0.5101
(0.4144)
0.5178
(0.3778)
-0.1528
(0.3627)
0.0998
(0.3376)
-0.1663
(0.3835)
0.0986
(0.3516)
-0.1355
(0.3659)
0.1206
(0.3418)
adjusted R-squared
Observations
Region Fixed-Effects
Controls
0.689
153
Yes
Simple
0.748
153
Yes
Rich
0.692
152
Yes
Simple
0.753
152
Yes
Rich
0.701
153
Yes
Simple
0.749
153
Yes
Rich
0.690
148
Yes
Simple
0.736
148
Yes
Rich
0.687
147
Yes
Simple
0.731
147
Yes
Rich
0.687
148
Yes
Simple
0.7354
148
Yes
Rich
Both panels report cross-country OLS estimates. In all specifications we drop countries with just one ethnic group (in (1)-(6)) or just one linguistic group (in (7)-(12)). The dependent
variable is the log of real GDP per capita in 2000. In all specifications in Panel A we control for the overall degree of spatial inequality in a country using the Gini coefficient of
lights per capita based on Thiessen polygons that use as input points the centroids of the linguistic homelands according to the Ethnologue dataset. In all specifications in Panel B we
control for the overall degree of spatial inequality in a country using the Gini coefficient of lights per capita based on polygons that have the same size (2.5 x 2.5 degrees).
In columns (1)-(6) we use the digitized version of the Atlas Narodov Mira (GREG) to aggregate lights per capita across ethnic homelands and construct the ethnic inequality
measures. In columns (7)-(12) we use the digitized version of the Ethnologue database to aggregate lights per capita across linguistic homelands and construct the ethnic inequality
measures. For the construction of the ethnic inequality measures (Gini coefficients) in columns (3), (4), (9), and (10) we exclude ethnic areas where capital cities fall. For the
construction of the ethnic inequality measures (Gini coefficients) in columns (5), (6), (11), and (12) we exclude small ethnic groups consisting of less than one percept of country’s
population. Odd-numbered columns include as controls absolute latitude, log land area, and log population in 2000 (simple set of controls). Even-numbered columns include also
control for an index of terrain ruggedness, the percentage of each country with fertile soil, the percentage of each country with tropical climate, average distance to nearest ice-free
coast, and an index of gem-quality diamond extraction (rich set of controls). In all specifications we control for ethnic fragmentation using an index that reflects the likelihood that
two randomly chosen individuals in one country will be members of the same group.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted
standard errors are reported in parentheses below the estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 3A: Geography, History and Contemporary Ethnic Inequality, Atlas Naodov Mira
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Log Land Area
0.0247**
(0.0098)
0.0235** 0.0204** 0.0071
(0.0096) (0.0094) (0.0152)
0.0147
(0.0101)
0.0171
(0.0111)
0.0034
(0.0124)
0.0024
(0.0113)
Log Population
0.003
(0.0117)
0.0029
(0.0127)
0.0004 -0.0046
(0.0125) (0.0173)
-0.0030
(0.0148)
-0.0052
(0.0149)
-0.0166
(0.0134)
-0.0164
(0.0212)
Latitude
0.0008
(0.0018)
0.0009
(0.0022)
0.0004 -0.0021
(0.0023) (0.0032)
0.0045**
(0.0022)
0.0030
(0.0022)
0.0026
(0.0023)
0.0053**
(0.0025)
Ruggedness
-0.0031
(0.0104)
-0.0051 0.0200
(0.0107) (0.0249)
(0.0022)
(0.0107)
0.0013
(0.0116)
0.0023
(0.0120)
0.0128
(0.0098)
Soil Quality
0.0002
(0.0006)
0.0003
0.0004
(0.0006) (0.0010)
0.0002
(0.0006)
-0.0003
(0.0007)
0.0005
(0.0007)
0.0002
(0.0008)
Tropical Climate
0.0001
(0.0005)
0.00001 -0.0003
(0.0005) (0.0006)
0.0007
(0.0005)
0.0009*
(0.0005)
0.0006
(0.0006)
0.0011**
(0.0006)
Gem Stones
0.0001
(0.0001)
0.0002
0.0001
(0.0003) (0.0002)
0.0002
(0.0004)
0.0003
(0.0002)
0.0001
(0.0002)
0.0002
(0.0005)
Distance to the Coast
0.0486
(0.0422)
0.0476
0.0677
(0.0412) (0.0606)
0.0226
(0.0437)
0.1407***
(0.0483)
0.0393
(0.0432)
0.0107
(0.0381)
Common Law
-0.0410
(0.0320)
Log Settler Mortality
0.0133
(0.0205)
European Descent
-0.0011*
(0.0006)
Executive Constraints
at Independence
-0.0012
(0.0411)
State Antiquity Index
0.1211
(0.0829)
Ethnic Partitioning
-0.0004
(0.0005)
Border Straightness
-0.0213
(0.8016)
Spatial Inequality 2
[Gini Coeff.]
0.6669*** 0.6515*** 0.6429***0.6650*** 0.7010*** 0.5805*** 0.7564*** 0.7448***
(0.0816) (0.0875) (0.0891) (0.1092) (0.0783) (0.0884) (0.0846)
(0.1047)
adjusted R-squared
Observations
Region Fixed-Effects
0.684
173
Yes
0.676
173
Yes
0.687
173
Yes
0.665
77
Yes
0.668
157
Yes
0.631
133
Yes
0.723
142
Yes
0.653
113
Yes
Appendix Table 3B: Geography, History and Contemporary Ethnic Inequality, Ethnologue
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Log Land Area
0.0247**
(0.0098)
0.0235** 0.0204** 0.0071
(0.0096) (0.0094) (0.0152)
0.0147
(0.0101)
0.0171
(0.0111)
0.0034
(0.0124)
0.0024
(0.0113)
Log Population
0.003
(0.0117)
0.0029
(0.0127)
0.0004 -0.0046
(0.0125) (0.0173)
-0.0030
(0.0148)
-0.0052
(0.0149)
-0.0166
(0.0134)
-0.0164
(0.0212)
Latitude
0.0008
(0.0018)
0.0009
(0.0022)
0.0004 -0.0021
(0.0023) (0.0032)
0.0045**
(0.0022)
0.0030
(0.0022)
0.0026
(0.0023)
0.0053**
(0.0025)
Ruggedness
-0.0031
(0.0104)
-0.0051 0.0200
(0.0107) (0.0249)
(0.0022)
(0.0107)
0.0013
(0.0116)
0.0023
(0.0120)
0.0128
(0.0098)
Soil Quality
0.0002
(0.0006)
0.0003
0.0004
(0.0006) (0.0010)
0.0002
(0.0006)
-0.0003
(0.0007)
0.0005
(0.0007)
0.0002
(0.0008)
Tropical Climate
0.0001
(0.0005)
0
-0.0003
(0.0005) (0.0006)
0.0007
(0.0005)
0.0009*
(0.0005)
0.0006
(0.0006)
0.0011**
(0.0006)
Gem Stones
0.0000
(0.0000)
0.0000
0.0000
(0.0000) (0.0000)
0.0000
(0.0000)
0.0000
(0.0000)
0.0000
(0.0000)
0.0000
(0.0000)
Distance to the Coast
0.0486
(0.0422)
0.0476
0.0677
(0.0412) (0.0606)
0.0226
(0.0437)
0.1407***
(0.0483)
0.0393
(0.0432)
0.0107
(0.0381)
Common Law
-0.0410
(0.0320)
Log Settler Mortality
0.0133
(0.0205)
European Descent
-0.0011*
(0.0006)
Executive Constraints
at Independence
State Antiquity Index
-0.0012
(0.0411)
0.1211
(0.0829)
Ethnic Partitioning
-0.0004
(0.0005)
-0.0213
(0.8016)
Border Straightness
Spatial Inequality 2
[Gini Coeff.]
adjusted R-squared
Observations
Region Fixed-Effects
0.6669*** 0.6515*** 0.6429***0.6650*** 0.7010*** 0.5805*** 0.7564*** 0.7448***
(0.0816) (0.0875) (0.0891) (0.1092) (0.0783) (0.0884) (0.0846)
(0.1047)
0.6652
173
Yes
0.659
173
Yes
0.6606
173
Yes
0.6239
77
Yes
0.6493
157
Yes
0.622
133
Yes
0.6826
142
Yes
0.6502
113
Yes
The table reports cross-country OLS estimates, associating contemporary ethnic inequality with various geographic and historical variables. The
dependent variable is the ethnic Gini coefficient that reflects inequality in lights per capita across ethnic-linguistic homelands, using the
digitized version of Atlas Narodov Mira (GREG) in Panel A and Ethnologue in Panel B. In all specifications we control for the overall degree of
spatial inequality in a country using the Gini coefficient of lights per capita based on Thiessen polygons that use as input points the centroids of
the linguistic homelands according to the Ethnologue dataset. All specifications include continental fixed effects (constants not reported). The
Data Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses
below the estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 4: Correlation Structure Inequality Measures in Geographic Endowments
Distance to the Sea
Elevation
Land Quality
Ethnic Gini - Sea Distance
Spatial Gini 1 - Sea Distance
Mean Sea Distance
1
0.7971*
-0.0306
1
-0.1910*
1
Ethnic Gini - Elevation
Spatial Gini 1 - Elevation
Mean Elevation
0.3879*
0.3313*
0.0254
0.2692*
0.2698*
-0.0979
0.2612*
0.2131*
0.5010*
1
0.8776*
0.6012*
1
0.5662*
1
Ethnic Gini - Land Quality
0.3702*
Spatial Gini 1 - Land Quality 0.3007*
Mean Land Quality
0.0299
0.2460*
0.2311*
0.0005
0.3698*
0.3577*
-0.1677*
0.5181*
0.4475*
-0.0075
0.4110*
0.4061*
0.0180
0.3367*
0.3360*
0.0238
1
0.9253*
1
-0.4825* -0.5423*
Ethnic Gini - Water Area
Spatial Gini 1 - Water Area
Mean Water Area
0.4833*
0.4924*
0.2538*
0.3505*
0.3819*
0.1472
0.5288*
0.4074*
0.2078*
0.3904*
0.3746*
0.1995*
0.3606*
0.3735*
0.0365
0.5002*
0.3315*
0.2996*
0.6298*
0.5081*
0.2944*
0.4133*
0.3049*
0.3041*
Water Area
1
0.1217
0.0735
-0.1798*
1
0.7775*
0.1648*
1
0.1236
1
The table reports the correlation structure between the main geographic variables employed in the cross-country analysis within African countries. Specifically the table gives the
correlation between inequality in geographic endowments across ethnic homelands, inequality in geographic endowments across pixels of 2.5x2.5 degrees, and the level of
geography, as reflected in distance to the sea, elevation, an index of land (soil) suitability (quality) for agriculture, and water area. * indicate statistical significance at the 5% level.
Appendix Table 5: The Origins of Contemporary Ethnic Inequality. Sensitivity Analysis
Inequality in Geographic Endowments across Ethnic Homelands and Contemporary Ethnic Inequality
Atlas Narodov Mira (GREG)
(1)
(2)
(3)
(4)
Ethnologue
(5)
(6)
Panel A: Excluding Capitals
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
0.0984*** 0.0846*** 0.0841***
(0.0102)
(0.0158)
(0.0147)
9.67
Spatial Inequality in Geographic
Endowments (PC)
adjusted R-squared
Observations
0.538
151
5.34
5.73
0.0256
(0.0209)
1.22
0.0122
(0.0211)
0.58
0.542
150
0.582
150
0.1422*** 0.1220*** 0.1334***
(0.0123) (0.0180)
(0.0175)
11.53
0.676
144
6.78
7.64
0.0282
(0.0224)
1.25
-0.0048
(0.0235)
-0.20
0.674
143
0.700
142
Panel B: Excluding Small Groups
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
0.0760*** 0.0710*** 0.0680***
(0.0072)
(0.0090)
(0.0090)
10.62
Spatial Inequality in Geographic
Endowments (PC)
adjusted R-squared
Observations
Region Fixed-Effects
Additional Controls
0.6165
173
Yes
No
7.92
7.57
0.0024
(0.0079)
0.31
0.0043
(0.0096)
0.45
0.6227
169
Yes
No
0.6106
162
Yes
Geography
0.0934*** 0.0957*** 0.0960***
(0.0079) (0.0105)
(0.0101)
11.76
0.6365
169
Yes
No
9.08
9.47
-0.0062
(0.0101)
-0.62
-0.0079
(0.0127)
-0.62
0.6348
167
Yes
No
0.6202
161
Yes
Geography
The table reports cross-country OLS estimates, associating contemporary ethnic inequality with inequality in geographic endowments
across ethnic homelands. The dependent variable is the ethnic Gini coefficient that reflects inequality in lights per capita across ethniclinguistic homelands in 2000, using the digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue (in
columns (4)-(6)). To construct the ethnic inequality index (Gini coefficient) we exclude ethnic regions where capital cities fall (in Panel
A) and ethnic regions where small ethnicities consisting less than one percent of a country’s population reside (in Panel B).
The main independent variable is a composite index capturing inequality in geographic endowments across ethnic homelands. The
index is the first principal component of inequality across ethnic-linguistic homelands in distance to the coast, elevation, land suitability
for agriculture, and area under water. In columns (2), (3), (5), and (6) we control for the overall degree of spatial inequality in
geographic endowments using a composite index that aggregates (via principal components) Gini coefficients on distance to the coast,
elevation, land suitability for agriculture, water area across Thiessen polygons that use as input points the centroids of the linguistic
homelands according to the Ethnologue dataset. In columns (3) and (6) we also control for the mean value of distance to the coast,
elevation, land suitability for agriculture, and area under water.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and
data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 6: Inequality in Geographic Endowments across Ethnic Homelands and Contemporary
Development. Sensitivity Analysis
Atlas Narodov Mira (GREG)
(1)
(2)
(3)
(4)
Ethnologue
(5)
(6)
Panel A: Excluding Capitals
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-0.1176**
(0.0476)
-0.1152*
(0.0676)
-0.1268*
(0.0704)
-0.0861
(0.0621)
-0.008
(0.0956)
-0.0593
(0.0980)
-2.47
-1.7
-1.80
-1.39
-0.08
-0.6
0.0052
(0.1094)
0.05
0.0153
(0.1244)
0.12
-0.1313
(0.1011)
-1.30
-0.0876
(0.1185)
-0.74
0.633
150
0.664
150
0.640
143
0.662
142
Spatial Inequality in Geographic
Endowments (PC)
adjusted R-squared
Observations
0.633
151
0.638
144
Panel B: Excluding Small Groups
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-0.2075*** -0.1951*** -0.1709***
(0.0431)
(0.0576)
(0.0591)
-4.81
-3.39
-2.89
Spatial Inequality in Geographic
Endowments (PC)
adjusted R-squared
Observations
Region Fixed Effects
Additional Controls
0.6397
173
Yes
No
-0.0061
(0.0557)
-0.11
0.0005
(0.0718)
0.01
0.6550
169
Yes
No
0.6780
162
Yes
Geography
-0.1781*** -0.1346*
(0.0446) (0.0695)
-4.00
-1.94
-0.1732**
(0.0671)
-2.58
-0.0267
(0.0735)
-0.36
0.0352
(0.0874)
0.40
0.6393
167
Yes
No
0.6729
161
Yes
Geography
0.6259
169
Yes
No
The table reports cross-country OLS estimates, associating contemporary economic development with inequality in geographic
endowments across ethnic homelands. The dependent variable is the log of real GDP per capita in 2000. To construct the ethnic
inequality index and the inequality in geographic endowments across ethnic homelands (Gini coefficients) we exclude ethnic regions
where capital cities fall (in Panel A) and ethnic regions where small ethnicities consisting less than one percent of a country’s
population reside (in Panel B).
The main independent variable is a composite index capturing inequality in geographic endowments across ethnic homelands, using the
digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue (in columns (4)-(6)). The index is the first
principal component of inequality across ethnic-linguistic homelands in distance to the coast, elevation, land suitability for agriculture,
and area under water. In columns (2), (3), (5), and (6) we control for the overall degree of spatial inequality in geographic endowments
using a composite index that aggregates (via principal components) Gini coefficients on distance to the coast, elevation, land suitability
for agriculture, water area across Thiessen polygons that use as input points the centroids of the linguistic homelands according to the
Ethnologue dataset. In columns (3) and (6) we also control for the mean value of distance to the coast, elevation, land suitability for
agriculture, and area under water.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions and
data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 7: Inequality in Geographic Endowments across Ethnic Homelands, Ethnic Inequality, and
Contemporary Development. 2SLS Estimates
Atlas Narodov Mira (GREG)
(1)
(2)
(3)
(4)
Ethnologue
(5)
(6)
Panel A: All Ethnic Homelands
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-1.4574*** -2.1223*** -1.8196**
(0.3034)
(0.7983)
(0.8996)
-4.8
-2.66
-2.02
Spatial Inequality in Geographic
Endowments (PC)
First-Stage F-score
Observations
195.96
173
0.0986
(0.1015)
0.97
0.0603
(0.1086)
0.56
37.54
169
32.43
162
-0.9983*** -0.3808
(0.2655) (0.6024)
-3.76
-0.63
-0.9613*
(0.5002)
-1.92
-0.0869
(0.0981)
-0.89
0.021
(0.0982)
0.21
24.99
166
26.43
160
138.19
168
Panel B: Excluding Capitals
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-1.1956**
(0.4778)
-1.3624*
(0.7382)
-1.5067*
(0.7919)
-0.6051
(0.4015)
-0.0657
(0.7518)
-0.4444
(0.6787)
-2.5
-1.85
-1.90
-1.51
-0.09
-0.65
0.04
(0.1146)
0.35
0.0337
(0.1228)
0.27
-0.1295
(0.1139)
-1.14
-0.0898
(0.1062)
-0.85
28.529
150
32.801
150
45.965
142
58.354
142
Spatial Inequality in Geographic
Endowments (PC)
First-Stage F-score
Observations
93.418
151
133.040
144
Panel C: Excluding Small Groups
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
-2.7288*** -2.7499*** -2.5126***
(0.5617)
(0.8320)
(0.8745)
-4.86
-3.31
-2.87
-1.9062*** -1.4074** -1.8042***
(0.4444) (0.7026)
(0.6425)
-4.29
-2.00
-2.81
0.0006
(0.0547)
0.01
0.0113
(0.0715)
0.16
-0.0355
(0.0668)
-0.53
0.0208
(0.0791)
0.26
112.78
173
62.693
169
57.242
162
138.19
169
82.427
167
89.658
161
Yes
No
Yes
No
Yes
Geography
Yes
No
Yes
No
Yes
Geography
Spatial Inequality in Geographic
Endowments (PC)
First-Stage F-score
Observations
Region Fixed Effects
Additional Controls
Table Notes
The table reports cross-country two-stage-least-squares (2SLS) estimates, associating contemporary inequality in lights per capita
across ethnic homelands with inequality in geographic endowments across ethnic homelands in the first stage and the component of
ethnic inequality explained by inequality in geographic endowments across ethnic homelands with economic development in the
second stage. The dependent variable in the second stage is the log of real GDP per capita in 2000. The dependent variable in the
first stage is the ethnic Gini coefficient that reflects inequality in lights per capita across ethnic-linguistic homelands in 2000, using
the digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue (in columns (4)-(6)). In Panel A we use
all ethnic-linguistic homelands. In Panel B we exclude ethnic-linguistic regions where capital cities fall. In Panel C we exclude ethniclinguisic regions where small ethnicities consisting less than one percent of a country’s population reside.
The main independent variable in the first stage is a composite index capturing inequality in geographic endowments across ethnic
homelands, using the digitized version of Atlas Narodov Mira (GREG) (in columns (1)-(3)) and Ethnologue (in columns (4)-(6)).
The index is the first principal component of inequality across ethnic-linguistic homelands in distance to the coast, elevation, land
suitability for agriculture, and area under water. In columns (2), (3), (5), and (6) we control for the overall degree of spatial inequality
in geographic endowments using a composite index that aggregates (via principal components) Gini coefficients on distance to the
coast, elevation, land suitability for agriculture, water area across Thiessen polygons that use as input points the centroids of the
linguistic homelands according to the Ethnologue dataset. In columns (3) and (6) we also control for the mean value of distance to
the coast, elevation, land suitability for agriculture, and area under water.
All specifications include continental fixed effects (constants not reported). The Data Appendix gives detailed variable definitions
and data sources. Heteroskedasticity-adjusted standard errors are reported in parentheses below the estimates. ***, **, and * indicate
statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 8: Inequality in Geographic Endowments across Ethnic Homelands, Ethnic Inequality, and
Contemporary Development. OLS Estimates
Contemporary Ethnic Inequality
Inequality in Geographic Endowments
across Ethnic Homelands (PC)
adjusted R-squared
Observations
Region Fixed Effects
Atlas Narodov Mira (GREG)
All
Excl.
Excl. Small
Homelands Capitals
Groups
(1)
(2)
(3)
All
Homelands
(4)
Ethnologue
Excl.
Excl. Small
Capitals
Groups
(5)
(6)
-1.4885*** -1.3292*** -1.6339***
(0.4017)
(0.3833)
(0.5041)
-3.71
-3.47
-3.24
-1.2844*** -1.5078*** -1.2004**
(0.3620)
(0.4006)
(0.4696)
-3.55
-3.76
-2.56
0.0038
(0.0617)
0.06
0.0131
(0.0561)
0.23
-0.0832
(0.0525)
-1.59
0.0462
(0.0680)
0.68
0.1284
(0.0841)
1.53
-0.066
(0.0582)
-1.13
0.659
173
Yes
0.659
151
Yes
0.663
173
Yes
0.652
168
Yes
0.678
144
Yes
0.641
169
Yes
The table reports cross-country OLS estimates, associating contemporary economic development with contemporary ethnic inequality
and inequality in geographic endowments across ethnic homelands. The dependent variable is the log of real GDP per capita in 2000.
In columns (1) and (4) we construct the ethnic inequality measures and the inequality in geographic endowments across ethnic
homelands (Gini coefficients) using all ethnic-linguistic homelands. In columns (2) and (5) we exclude ethnic-linguistic regions where
capital cities fall. In columns (3) and (6) we exclude ethnic-linguistic regions where small ethnicities consisting less than one percent of
a country’s population reside. The main independent variables are and index capturing contemporary differences in development (as
reflected in lights per capita in 2000) across ethnic homelands and a composite index capturing inequality in geographic endowments
across ethnic homelands. The index is the first principal component of inequality across ethnic-linguistic homelands in distance to the
coast, elevation, land suitability for agriculture, and area under water. In columns (1)-(3) we use the digitized version of Atlas Narodov
Mira (GREG) and in columns (4)-(6) we are using the Ethnologue maps. All specifications include continental fixed effects (constants
not reported). The Data Appendix gives detailed variable definitions and data sources. Heteroskedasticity-adjusted standard errors are
reported in parentheses below the estimates. ***, **, and * indicate statistical significance at the 1%, 5%, and 10% level, respectively.
Appendix Table 9: Correlation Structure Afrobarometer Data at the District Level
Ethnic Inequality Indicators
Theil Index - Overall
Theil Index - Between
Theil Index - Within
Mean Log Deviation - Overall
Mean Log Deviation - Between
Mean Log Deviation - Within
Living Conditions Index
Education
Sewage System
Clean Piped Water
Access to Electricity Grid
1
0.9077*
0.5687*
0.9893*
0.8985*
0.5568*
-0.2169*
-0.0114
-0.0230
-0.0098
-0.0231
1
0.1710*
0.8951*
0.9863*
0.1637*
-0.2095*
-0.0308
-0.0302
-0.0133
-0.0429
1
0.5683*
0.1762*
0.9864*
-0.0986*
0.0336
0.0053
0.0031
0.0299
1
0.9072*
0.5649*
-0.1603*
0.0196
-0.0076
0.0039
0.0010
1
0.1653*
1
-0.1518* -0.0781*
0.0019
0.0422
-0.0158 0.0132
-0.0020 0.0130
-0.0153 0.0324
Development Proxy Measures
1.0000
0.3983*
0.1696* 0.3944* 1.0000
0.1319* 0.2747* 0.5517*
1
0.2331* 0.4803* 0.5185* 0.5280*
1
The table reports the correlation structure between the main variables employed in the cross-region analysis within African countries (Afrobarometer Sample). The Data
Appendix gives detailed variable definitions and data sources. * indicate statistical significance at the 5% level.